⬆️(dependencies) update python dependencies

Signed-off-by: Laurent Paoletti <lp@providenz.fr>

.codespellrc

@@ -0,0 +1,20 @@
+[codespell]
+skip =
+    ./git/,
+    **/*.pdf,
+    **/*.po,
+    **/*.pot,
+    **/*.json,
+    **/yarn.lock,
+    **/node_modules/**,
+    **/e2e/report/**,
+    *.tsbuildinfo,
+    **/uv.lock,
+    ./docker/files/etc/mime.types,
+check-filenames = true
+ignore-words-list =
+    afterAll,
+    statics,
+exclude-file =
+    ./src/backend/chat/agent_rag/web_search/mocked.py,
+

.github/ISSUE_TEMPLATE/Bug_report.md

@@ -18,7 +18,7 @@ A clear and concise description of what you expected to happen (or code).
 3. And then the bug happens!
 
 **Environment**
-- Docs version:
+- Conversations version:
 - Instance url:
 
 **Possible Solution**

.github/workflows/conversations-frontend.yml

@@ -18,7 +18,7 @@ jobs:
 
   test-front:
     needs: install-dependencies
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
@@ -28,6 +28,9 @@ jobs:
         with:
           node-version: "22.x"
 
+      - name: Install yarn
+        run: npm install -g yarn
+
       - name: Restore the frontend cache
         uses: actions/cache@v4
         with:
@@ -39,7 +42,7 @@ jobs:
         run: cd src/frontend/ && yarn test
 
   lint-front:
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     needs: install-dependencies
     steps:
       - name: Checkout repository
@@ -49,6 +52,10 @@ jobs:
         uses: actions/setup-node@v4
         with:
           node-version: "22.x"
+
+      - name: Install yarn
+        run: npm install -g yarn
+
       - name: Restore the frontend cache
         uses: actions/cache@v4
         with:
@@ -60,9 +67,9 @@ jobs:
         run: cd src/frontend/ && yarn lint
 
   test-e2e-chromium:
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     needs: install-dependencies
-    timeout-minutes: 20
+    timeout-minutes: 40
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
@@ -72,6 +79,9 @@ jobs:
         with:
           node-version: "22.x"
 
+      - name: Install yarn
+        run: npm install -g yarn
+
       - name: Restore the frontend cache
         uses: actions/cache@v4
         with:
@@ -99,9 +109,9 @@ jobs:
           retention-days: 7
 
   test-e2e-other-browser:
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     needs: test-e2e-chromium
-    timeout-minutes: 20
+    timeout-minutes: 40
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
@@ -111,6 +121,9 @@ jobs:
         with:
           node-version: "22.x"
 
+      - name: Install yarn
+        run: npm install -g yarn
+
       - name: Restore the frontend cache
         uses: actions/cache@v4
         with:

.github/workflows/conversations.yml

@@ -15,7 +15,7 @@ jobs:
       with-build_mails: true
 
   lint-git:
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     if: github.event_name == 'pull_request' # Makes sense only for pull requests
     steps:
       - name: Checkout repository
@@ -25,18 +25,22 @@ jobs:
       - name: show
         run: git log
       - name: Enforce absence of print statements in code
+        if: always()
         run: |
           ! git diff origin/${{ github.event.pull_request.base.ref }}..HEAD -- . ':(exclude)**/conversations.yml' | grep "print("
       - name: Check absence of fixup commits
+        if: always()
         run: |
           ! git log | grep 'fixup!'
       - name: Install gitlint
+        if: always()
         run: pip install --user requests gitlint
       - name: Lint commit messages added to main
+        if: always()
         run: ~/.local/bin/gitlint --commits origin/${{ github.event.pull_request.base.ref }}..HEAD
 
   check-changelog:
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     if: |
       contains(github.event.pull_request.labels.*.name, 'noChangeLog') == false &&
       github.event_name == 'pull_request'
@@ -49,7 +53,7 @@ jobs:
         run: git diff --name-only ${{ github.event.pull_request.base.sha }} ${{ github.event.after }} | grep 'CHANGELOG.md'
 
   lint-changelog:
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     steps:
       - name: Checkout repository
         uses: actions/checkout@v2
@@ -62,7 +66,7 @@ jobs:
           fi
 
   lint-spell-mistakes:
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     if: github.event_name == 'pull_request'
     steps:
       - name: Checkout repository
@@ -70,18 +74,10 @@ jobs:
       - name: Install codespell
         run: pip install --user codespell
       - name: Check for typos
-        run: |
-          codespell \
-            --check-filenames \
-            --ignore-words-list "Dokument,afterAll,excpt,statics" \
-            --skip "./git/" \
-            --skip "**/*.po" \
-            --skip "**/*.pot" \
-            --skip "**/*.json" \
-            --skip "**/yarn.lock"
+        run: codespell
 
   lint-back:
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     defaults:
       run:
         working-directory: src/backend
@@ -89,22 +85,27 @@ jobs:
       - name: Checkout repository
         uses: actions/checkout@v2
       - name: Install Python
-        uses: actions/setup-python@v3
+        uses: actions/setup-python@v6
         with:
-          python-version: "3.13.3"
-      - name: Upgrade pip and setuptools
-        run: pip install --upgrade pip setuptools
-      - name: Install development dependencies
-        run: pip install --user .[dev]
+          python-version: "3.13"
+      - name: Install system dependencies for lxml
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y libxml2-dev libxslt-dev
+      - name: Install uv
+        uses: astral-sh/setup-uv@v6
+      - name: Install the project
+        run: uv sync --locked --all-extras
+
       - name: Check code formatting with ruff
-        run: ~/.local/bin/ruff format . --diff
+        run: uv run ruff format . --diff
       - name: Lint code with ruff
-        run: ~/.local/bin/ruff check .
+        run: uv run ruff check .
       - name: Lint code with pylint
-        run: ~/.local/bin/pylint .
+        run: uv run pylint .
 
   test-back:
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     needs: install-dependencies
 
     defaults:
@@ -184,21 +185,48 @@ jobs:
             mc version enable conversations/conversations-media-storage"
 
       - name: Install Python
-        uses: actions/setup-python@v3
+        uses: actions/setup-python@v6
         with:
-          python-version: "3.13.3"
-
-      - name: Install development dependencies
-        run: pip install --user .[dev]
+          python-version: "3.13"
+      - name: Install system dependencies for lxml
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y libxml2-dev libxslt-dev
+      - name: Install uv
+        uses: astral-sh/setup-uv@v6
+      - name: Install the dependencies
+        run: uv sync --locked --all-extras
 
       - name: Install gettext (required to compile messages) and MIME support
         run: |
-          sudo apt-get update
           sudo apt-get install -y gettext pandoc shared-mime-info
-          sudo wget https://svn.apache.org/repos/asf/httpd/httpd/trunk/docs/conf/mime.types -O /etc/mime.types
+          sudo cp $GITHUB_WORKSPACE/docker/files/etc/mime.types /etc/mime.types
 
       - name: Generate a MO file from strings extracted from the project
-        run: python manage.py compilemessages
+        run: uv run python manage.py compilemessages
 
       - name: Run tests
-        run: ~/.local/bin/pytest -n 2
+        run: uv run pytest -n 2
+
+  security-trivy-critical:
+    permissions:
+      contents: read
+      security-events: write
+    runs-on: ubuntu-latest
+    steps:
+      - name: Run Trivy analysis for critical vulnerabilities
+        # We use main branch while we might still iterate on the action
+        uses: numerique-gouv/action-trivy-cache/security-trivy-critical@main
+        with:
+          skip-files: src/mail/yarn.lock
+
+  security-trivy:
+    permissions:
+      contents: read
+    runs-on: ubuntu-latest
+    steps:
+      - name: Run Trivy analysis for vulnerabilities
+        # We use main branch while we might still iterate on the action
+        uses: numerique-gouv/action-trivy-cache/security-trivy@main
+        with:
+          skip-files: src/mail/yarn.lock

.github/workflows/crowdin_download.yml

@@ -14,7 +14,7 @@ jobs:
       with-front-dependencies-installation: true
 
   synchronize-with-crowdin:
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     permissions:
       contents: write
       pull-requests: write
@@ -47,6 +47,12 @@ jobs:
 
           CROWDIN_BASE_PATH: "../src/"
       # frontend i18n
+      - name: Setup Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: "22.x"
+      - name: Install yarn
+        run: npm install -g yarn
       - name: Restore the frontend cache
         uses: actions/cache@v4
         with:
@@ -74,4 +80,4 @@ jobs:
 
             - [x]  update translated strings
           branch: i18n/update-translations
-          labels: i18n
+          labels: i18n,noChangeLog

.github/workflows/crowdin_upload.yml

@@ -16,20 +16,20 @@ jobs:
 
   synchronize-with-crowdin:
     needs: install-dependencies
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
 
     steps:
       - name: Checkout
         uses: actions/checkout@v4
       # Backend i18n
-      - name: Install Python
-        uses: actions/setup-python@v3
+      - name: Set up Python
+        uses: actions/setup-python@v6
         with:
-          python-version: "3.13.3"
-      - name: Upgrade pip and setuptools
-        run: pip install --upgrade pip setuptools
-      - name: Install development dependencies
-        run: pip install --user .
+          python-version-file: "src/backend/pyproject.toml"
+      - name: Install uv
+        uses: astral-sh/setup-uv@v6
+      - name: Install the project
+        run: uv sync --locked --all-extras
         working-directory: src/backend
       - name: Restore the mail templates
         uses: actions/cache@v4
@@ -45,8 +45,14 @@ jobs:
       - name: generate pot files
         working-directory: src/backend
         run: |
-          DJANGO_CONFIGURATION=Build python manage.py makemessages -a --keep-pot
+          DJANGO_CONFIGURATION=Build uv run python manage.py makemessages -a --keep-pot
       # frontend i18n
+      - name: Setup Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: "22.x"
+      - name: Install yarn
+        run: npm install -g yarn
       - name: Restore the frontend cache
         uses: actions/cache@v4
         with:

.github/workflows/dependencies.yml

@@ -17,24 +17,32 @@ on:
 jobs:
   front-dependencies-installation:
     if: ${{ inputs.with-front-dependencies-installation == true }}
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     steps:
       - name: Checkout
         uses: actions/checkout@v4
+
       - name: Restore the frontend cache
         uses: actions/cache@v4
         id: front-node_modules
         with:
           path: "src/frontend/**/node_modules"
           key: front-node_modules-${{ hashFiles('src/frontend/**/yarn.lock') }}
+
       - name: Setup Node.js
         if: steps.front-node_modules.outputs.cache-hit != 'true'
         uses: actions/setup-node@v4
         with:
           node-version: ${{ inputs.node_version }}
+
+      - name: Install yarn
+        if: steps.front-node_modules.outputs.cache-hit != 'true'
+        run: npm install -g yarn
+
       - name: Install dependencies
         if: steps.front-node_modules.outputs.cache-hit != 'true'
         run: cd src/frontend/ && yarn install --frozen-lockfile
+
       - name: Cache install frontend
         if: steps.front-node_modules.outputs.cache-hit != 'true'
         uses: actions/cache@v4
@@ -44,7 +52,7 @@ jobs:
 
   build-mails:
     if: ${{ inputs.with-build_mails == true }}
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     defaults:
       run:
         working-directory: src/mail

.github/workflows/docker-hub.yml

@@ -18,11 +18,14 @@ env:
 
 jobs:
   build-and-push-backend:
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     steps:
       -
         name: Checkout repository
         uses: actions/checkout@v4
+      -
+        name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
       -
         name: Docker meta
         id: meta
@@ -32,7 +35,10 @@ jobs:
       -
         name: Login to DockerHub
         if: github.event_name != 'pull_request'
-        run: echo "${{ secrets.DOCKER_HUB_PASSWORD }}" | docker login -u "${{ secrets.DOCKER_HUB_USER }}" --password-stdin
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKER_HUB_USER }}
+          password: ${{ secrets.DOCKER_HUB_PASSWORD }}
       -
         name: Run trivy scan
         uses: numerique-gouv/action-trivy-cache@main
@@ -41,6 +47,7 @@ jobs:
           docker-image-name: 'docker.io/lasuite/conversations-backend:${{ github.sha }}'
       -
         name: Build and push
+        if: always()
         uses: docker/build-push-action@v6
         with:
           context: .
@@ -51,11 +58,14 @@ jobs:
           labels: ${{ steps.meta.outputs.labels }}
 
   build-and-push-frontend:
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     steps:
       -
         name: Checkout repository
         uses: actions/checkout@v4
+      -
+        name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
       -
         name: Docker meta
         id: meta
@@ -65,7 +75,10 @@ jobs:
       -
         name: Login to DockerHub
         if: github.event_name != 'pull_request'
-        run: echo "${{ secrets.DOCKER_HUB_PASSWORD }}" | docker login -u "${{ secrets.DOCKER_HUB_USER }}" --password-stdin
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKER_HUB_USER }}
+          password: ${{ secrets.DOCKER_HUB_PASSWORD }}
       -
         name: Run trivy scan
         uses: numerique-gouv/action-trivy-cache@main
@@ -74,6 +87,7 @@ jobs:
           docker-image-name: 'docker.io/lasuite/conversations-frontend:${{ github.sha }}'
       -
         name: Build and push
+        if: always()
         uses: docker/build-push-action@v6
         with:
           context: .
@@ -89,7 +103,7 @@ jobs:
     needs:
       - build-and-push-frontend
       - build-and-push-backend
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     if: github.event_name != 'pull_request'
     steps:
       - uses: numerique-gouv/action-argocd-webhook-notification@main

.github/workflows/helmfile-linter.yaml

@@ -9,7 +9,7 @@ on:
 
 jobs:
   helmfile-lint:
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     container:
       image: ghcr.io/helmfile/helmfile:v0.171.0
     steps:
@@ -21,10 +21,10 @@ jobs:
       shell: bash
       run: |
         set -e
-        HELMFILE=src/helm/helmfile.yaml
+        HELMFILE=src/helm/helmfile.yaml.gotmpl
         environments=$(awk 'BEGIN {in_env=0} /^environments:/ {in_env=1; next} /^---/ {in_env=0} in_env && /^  [^ ]/ {gsub(/^  /,""); gsub(/:.*$/,""); print}' "$HELMFILE")
         for env in $environments; do
           echo "################### $env lint ###################"
           helmfile -e $env -f $HELMFILE lint || exit 1
           echo -e "\n"
-        done
+        done

.github/workflows/release-helm-chart.yaml

@@ -12,7 +12,7 @@ jobs:
     # see: https://docs.github.com/en/actions/security-guides/automatic-token-authentication#modifying-the-permissions-for-the-github_token
     permissions:
       contents: write
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     steps:
       - name: Checkout
         uses: actions/checkout@v4

.gitignore

@@ -44,6 +44,9 @@ env.d/development/*
 !env.d/development/*.dist
 env.d/terraform
 
+# Configuration
+**/conversations/configuration/llm/dev.json
+
 # npm
 node_modules
 
@@ -76,3 +79,6 @@ db.sqlite3
 .vscode/
 *.iml
 .devcontainer
+
+# Docker compose override
+compose.override.yml

CHANGELOG.md

@@ -8,5 +8,256 @@ and this project adheres to
 
 ## [Unreleased]
 
+### Added
 
-[unreleased]: https://github.com/numerique-gouv/conversations/compare/HEAD...main
+- ✨(back) add projects with custom LLM instructions
+- ✨(front) projects management UI
+
+## [0.0.14] - 2026-03-11
+
+### Added
+
+- ✨(user) allow disabling automatic internet search
+- ✨(search) add searchModal and modify leftPanel
+- ✨(waffle) hide the waffle if not fr theme
+- ✨(front) allow pasting an attachment from clipboard
+- ✨(array) temporarily adjust array
+
+### Changed
+
+- ⚡️(front) optimize streaming markdown rendering performance
+- ⬆️(back) update pydantic-ai
+- ♻️(chat) refactor AIAgentService for readability and maintainability
+- 🚸(oidc) ignore case when fallback on email #281
+- ⬆️(back) update pillow, django-pydantic-field, pypdf
+- ♻️(front) migrate from ESLint 8 to ESLint 9 flat config
+- ⬆️(back) update django and pypdf
+
+### Fixed
+
+- 💚(docker) vendor mime.types file instead of fetching from Apache SVN
+- 🚑️(back) fix mime type for pptx
+- 🐛(front) fix math formulas and carousel translations
+- 🐛(helm) reverse liveness and readiness for backend deployment
+- 🐛(front) fix dark mode styling on chat messages
+- 🐛(front) fixed inverted toast for setting changes
+
+## [0.0.13] - 2026-02-09
+
+### Added
+
+- 💄(front) ui fix : update ui-kit
+- ✨(front) add persistent darkmode
+- ✨(front) add ui kit #240
+- 🧱(files) allow to use S3 storage without external access #849
+- ✨(backend) add FindRagBackend #209
+- ⬆️(back) update dependencies
+- ✨(back) use adaptive parsing for pdf documents
+
+### Changed
+
+- 💄(darkmode) change color feedback button
+- 🏗️(back) migrate to uv
+- ♻️(front) optimize syntax highlighting bundle size
+
+### Fixed
+
+- 🐛(back) cast collection Ids to API expected types
+
+## [0.0.12] - 2026-01-27
+
+### Fixed
+
+- ⚡️(front) performance improvements on chat input
+- 💄(front) i18n and standardize pdf parsing display
+
+### Removed
+
+- 🔥(chat) consider PDF documents as other kind of documents #234
+
+## [0.0.11] - 2026-01-16
+
+### Changed
+
+- 📦️(front) update react
+- ✨(chat) generate and edit conversation title
+
+### Fixed
+
+- 🐛(e2e) fix test-e2e-chromium
+- 🐛(back) fix system prompt compatibility with self-hosted models #200
+- ⚰️(back) remove dead code and unused files
+- 🐛(back) prevent tool call timeouts
+
+### Removed
+
+- 🔥(chat) remove thinking part from frontend #227
+
+## [0.0.10] - 2025-12-15
+
+### Added
+
+- ✨(front) add retry button
+
+### Fixed
+
+- 🐛(front) fix long user messages
+- 🐛(front) fix "Maximum update depth exceeded" error in Chat component
+- 🐛(front) fix parsing documents display
+- 🐛(front) fix opacity input in error
+- 🐛(front) resolve React hydration errors
+- 🚑️(user) allow longer short names #182
+
+## [0.0.9] - 2025-11-17
+
+### Added
+
+- ✨(front) add code copy button
+- ✨(RAG) add generic collection RAG tools #159
+
+### Fixed
+
+- 🔊(langfuse) enable tracing with redacted content #162
+
+## [0.0.8] - 2025-11-10
+
+### Fixed
+
+- 🦺(front) Fix send prohibited file types
+- 🐛(front) fix target blank links in chat #103
+- 🚑️(posthog) pass str instead of UUID for user PK #134
+- ⚡️(web-search) keep running when tool call fails #137
+- ✨(summarize): new summarize tool integration #78
+
+### Removed
+
+- 🔥(posthog) remove posthog middleware for async mode fix #146
+
+## [0.0.7] - 2025-10-28
+
+### Fixed
+
+- 🚑️(posthog) fix the posthog middleware for async mode #133
+
+## [0.0.6] - 2025-10-28
+
+### Fixed
+
+- 🚑️(stats) fix tracking id in upload event #130
+
+## [0.0.5] - 2025-10-27
+
+### Fixed
+
+- 🚑️(drag-drop) fix the rejection display on Safari #127
+
+## [0.0.4] - 2025-10-27
+
+### Added
+
+- ♿️(a11y) improve accessibility #135
+- 🌐(i18n) add dutch language #117
+
+### Changed
+
+- ⚡️(asgi) use `uvicorn` to serve backend #121
+
+### Fixed
+
+- 🐛(front) fix mobile source
+- 🐛(attachments) reject the whole drag&drop if unsupported formats #123
+
+## [0.0.3] - 2025-10-21
+
+### Fixed
+
+- 🚑️(web-search) fix missing argument in RAG backend #116
+
+## [0.0.2] - 2025-10-21
+
+### Added
+
+- ✨(front) add drag'n drop file
+- ✨(activation-codes) register users also on Brevo #98
+- 📈(posthog) add `sub` field to tracking #95
+
+### Changed
+
+- 🔧(front) change links feedback tchap + settings popup
+- 🐛(front) code activation fix session end #93
+- 💬(wording) error page wording #102
+- ⚡️(web-search) allow to override returned chunks #107
+- 🐛(activation-codes) create contact in brevo before add to list #108
+- ⚗️(summarization) add system prompt to handle tool #112
+
+## [0.0.1] - 2025-10-19
+
+### Changed
+
+- 🎨(front) activation page footer
+- 👷(front) change size small modal
+- 🎨(front) retour ui global
+- 👷(front) fix button scrollDown
+- 💥(front) disable input when error occurred
+- 👷(front) fix scroll
+- 🐛(front) fix left panel status + fix scroll
+- 🐛(llm) add is_active field and persist chat preference
+- ✨(frontend) add LLM selection in chat input #53
+- 🎨(front) fix width chat container
+- 🎨(front) fix width chat container #55
+- 🐛(front) fix button search web on new conversation
+- 🎨(front) improvement search input scroll
+- ✨(404) fix front 404 page
+- ✅(chat) add frontend feature flags #29
+- 🎨(front) change list attachment in chat
+- 🎨(front) move emplacement for attachment
+- 🎨(ui) retour ui sources files
+- ✨(ui) fix retour global ui
+- 🐛(fix) broken staging css
+- 🎨(alpha) adjustment for alpha version
+- ✨(ui) delete flex message
+- ✅(front) add enabled/disabled conversation analysis
+- 🎨(front) amelioration chat ux
+- 🎨(front) global layout modification
+- ✨(front) global layout UI
+- ♻️(chat) rewrite backend using Pydantic AI SDK #4
+- 🗃️(chat) enforce messages stored JSON format #6
+- 🐛(chat) UI messages must have a unique identifier #6
+- ✨(llm) allow configuration from JSON file #22
+- 💥(agent) replace routing w/ tool calls #40
+- 🧱(storage) upload the user documents into S3 #86
+
+### Added
+
+- 🎉(conversations) bootstrap backend & frontend #1
+- ✨(web-search) add RAG capability to do web search #7
+- ✨(chat) add document RAG on document uploaded by user #8
+- ✨(backend) allow use to stop conversation streaming #14
+- 🐛(agent) add the current date in the system prompt #18
+- ✨(backend) add feature flags from posthog #13
+- ✨(user) allow to use conversation data for analytics #23
+- ✨(chat) enforce response in user language #24
+- 📈(langfuse) add light instrumentation #26
+- 🚑️(agent) allow Mistral w/ vLLM & tools #36
+- ✨(web-search) add Brave search tool #47
+- ✨(models) add mistral support & customization #51
+- 🐛(web-search) add summarization to Brave results #58
+- ✨(langfuse) allow user to score messages from LLM #6
+- ✨(onboarding) add activation code logic for launch #62
+- 💄(chat) add code highlighting for LLM responses #67
+
+[unreleased]: https://github.com/suitenumerique/conversations/compare/v0.0.14...main
+[0.0.14]: https://github.com/suitenumerique/conversations/releases/v0.0.14
+[0.0.13]: https://github.com/suitenumerique/conversations/releases/v0.0.13
+[0.0.12]: https://github.com/suitenumerique/conversations/releases/v0.0.12
+[0.0.11]: https://github.com/suitenumerique/conversations/releases/v0.0.11
+[0.0.10]: https://github.com/suitenumerique/conversations/releases/v0.0.10
+[0.0.9]: https://github.com/suitenumerique/conversations/releases/v0.0.9
+[0.0.8]: https://github.com/suitenumerique/conversations/releases/v0.0.8
+[0.0.7]: https://github.com/suitenumerique/conversations/releases/v0.0.7
+[0.0.6]: https://github.com/suitenumerique/conversations/releases/v0.0.6
+[0.0.5]: https://github.com/suitenumerique/conversations/releases/v0.0.5
+[0.0.4]: https://github.com/suitenumerique/conversations/releases/v0.0.4
+[0.0.3]: https://github.com/suitenumerique/conversations/releases/v0.0.3
+[0.0.2]: https://github.com/suitenumerique/conversations/releases/v0.0.2
+[0.0.1]: https://github.com/suitenumerique/conversations/releases/v0.0.1

CONTRIBUTING.md

@@ -2,7 +2,7 @@
 
 Thank you for taking the time to contribute! Please follow these guidelines to ensure a smooth and productive workflow. 🚀🚀🚀
 
-To get started with the project, please refer to the [README.md](https://github.com/suitenumerique/conversations/blob/main/README.md) for detailed instructions on how to run Docs locally.
+To get started with the project, please refer to the [README.md](https://github.com/suitenumerique/conversations/blob/main/README.md) for detailed instructions on how to run Conversations locally.
 
 Contributors are required to sign off their commits with `git commit --signoff`: this confirms that they have read and accepted the [Developer's Certificate of Origin 1.1](https://developercertificate.org/). For security reasons we also require [signing your commits with your SSH or GPG key](https://docs.github.com/en/authentication/managing-commit-signature-verification/about-commit-signature-verification) with `git commit -S`.
 
@@ -92,11 +92,3 @@ Make sure that all new features or fixes have corresponding tests. Run the test
 If you need any help while contributing, feel free to open a discussion or ask for guidance in the issue tracker. We are more than happy to assist!
 
 Thank you for your contributions! 👍
-
-## Contribute to BlockNote
-We use [BlockNote](https://www.blocknotejs.org/) for the text editing features of Docs. 
-If you find and issue with the editor you can [report it](https://github.com/TypeCellOS/BlockNote/issues) directly on their repository.
-
-Please consider contributing to BlockNotejs, as a library, it's useful to many projects not just Docs.
-
-The project is licended with Mozilla Public License Version 2.0 but be aware that [XL packages](https://github.com/TypeCellOS/BlockNote/blob/main/packages/xl-docx-exporter/LICENSE) are dual licenced with GNU AFFERO GENERAL PUBLIC LICENCE Version 3 and proprietary licence if you are [sponsor](https://www.blocknotejs.org/pricing).

Dockerfile

@@ -3,9 +3,6 @@
 # ---- base image to inherit from ----
 FROM python:3.13.3-alpine AS base
 
-# Upgrade pip to its latest release to speed up dependencies installation
-RUN python -m pip install --upgrade pip setuptools
-
 # Upgrade system packages to install security updates
 RUN apk update && \
   apk upgrade
@@ -13,21 +10,31 @@ RUN apk update && \
 # ---- Back-end builder image ----
 FROM base AS back-builder
 
-WORKDIR /builder
+
+ENV UV_COMPILE_BYTECODE=1
+ENV UV_LINK_MODE=copy
+ENV UV_PYTHON_DOWNLOADS=0
+
+COPY --from=ghcr.io/astral-sh/uv:0.9.26 /uv /uvx /bin/
 
 # Install Rust and Cargo using Alpine's package manager
 RUN apk add --no-cache \
   build-base \
   libffi-dev \
+  libxml2-dev \
+  libxslt-dev \
   rust \
   cargo
 
-# Copy required python dependencies
-COPY ./src/backend /builder
-
-RUN mkdir /install && \
-  pip install --prefix=/install .
+WORKDIR /app
 
+RUN --mount=type=cache,target=/root/.cache/uv \
+    --mount=type=bind,source=src/backend/uv.lock,target=uv.lock \
+    --mount=type=bind,source=src/backend/pyproject.toml,target=pyproject.toml \
+    uv sync --locked --no-install-project --no-dev
+COPY src/backend /app
+RUN --mount=type=cache,target=/root/.cache/uv \
+    uv sync --locked --no-dev
 
 # ---- mails ----
 FROM node:24 AS mail-builder
@@ -49,14 +56,16 @@ RUN apk add \
   pango \
   rdfind
 
-# Copy installed python dependencies
-COPY --from=back-builder /install /usr/local
+WORKDIR /app
+
+# Copy the application from the builder
+COPY --from=back-builder /app /app
+
+ENV PATH="/app/.venv/bin:$PATH"
 
 # Copy conversations application (see .dockerignore)
 COPY ./src/backend /app/
 
-WORKDIR /app
-
 # collectstatic
 RUN DJANGO_CONFIGURATION=Build \
     python manage.py collectstatic --noinput
@@ -79,10 +88,12 @@ RUN apk add \
   gettext \
   gdk-pixbuf \
   libffi-dev \
+  libxml2 \
+  libxslt \
   pango \
   shared-mime-info
 
-RUN wget https://svn.apache.org/repos/asf/httpd/httpd/trunk/docs/conf/mime.types -O /etc/mime.types
+COPY ./docker/files/etc/mime.types /etc/mime.types
 
 # Copy entrypoint
 COPY ./docker/files/usr/local/bin/entrypoint /usr/local/bin/entrypoint
@@ -92,17 +103,17 @@ COPY ./docker/files/usr/local/bin/entrypoint /usr/local/bin/entrypoint
 # docker user (see entrypoint).
 RUN chmod g=u /etc/passwd
 
-# Copy installed python dependencies
-COPY --from=back-builder /install /usr/local
 
-# Copy conversations application (see .dockerignore)
-COPY ./src/backend /app/
+# Copy the application from the builder
+COPY --from=back-builder /app /app
 
 WORKDIR /app
 
+ENV PATH="/app/.venv/bin:$PATH"
+
 # Generate compiled translation messages
 RUN DJANGO_CONFIGURATION=Build \
-    python manage.py compilemessages
+    python manage.py compilemessages --ignore=".venv/**/*"
 
 
 # We wrap commands run in this container by the following entrypoint that
@@ -119,10 +130,9 @@ USER root:root
 # Install psql
 RUN apk add postgresql-client
 
-# Uninstall conversations and re-install it in editable mode along with development
-# dependencies
-RUN pip uninstall -y conversations
-RUN pip install -e .[dev]
+ # Install development dependencies
+RUN --mount=from=ghcr.io/astral-sh/uv:0.9.26,source=/uv,target=/bin/uv \
+    uv sync --all-extras --locked
 
 # Restore the un-privileged user running the application
 ARG DOCKER_USER
@@ -144,7 +154,7 @@ RUN rm -rf /var/cache/apk/*
 
 ARG CONVERSATIONS_STATIC_ROOT=/data/static
 
-# Gunicorn
+# Gunicorn - not used by default but configuration file is provided
 RUN mkdir -p /usr/local/etc/gunicorn
 COPY docker/files/usr/local/etc/gunicorn/conversations.py /usr/local/etc/gunicorn/conversations.py
 
@@ -158,5 +168,18 @@ COPY --from=link-collector ${CONVERSATIONS_STATIC_ROOT} ${CONVERSATIONS_STATIC_R
 # Copy conversations mails
 COPY --from=mail-builder /mail/backend/core/templates/mail /app/core/templates/mail
 
-# The default command runs gunicorn WSGI server in conversations's main module
-CMD ["gunicorn", "-c", "/usr/local/etc/gunicorn/conversations.py", "conversations.wsgi:application"]
+# The default command runs uvicorn ASGI server in conversations's main module
+# WEB_CONCURRENCY: number of workers to run <=> --workers=4
+ENV WEB_CONCURRENCY=4
+CMD [\
+    "uvicorn",\
+    "--app-dir=/app",\
+    "--host=0.0.0.0",\
+    "--timeout-graceful-shutdown=300",\
+    "--limit-max-requests=20000",\
+    "--lifespan=off",\
+    "conversations.asgi:application"\
+]
+
+# To run using gunicorn WSGI server use this instead:
+#CMD ["gunicorn", "-c", "/usr/local/etc/gunicorn/conversations.py", "conversations.wsgi:application"]

Makefile

@@ -126,7 +126,7 @@ build-frontend: ## build the frontend container
 build-e2e: cache ?=
 build-e2e: ## build the e2e container
 	@$(MAKE) build-backend cache=$(cache)
-	@$(COMPOSE_E2E) build frontend $(cache)
+	@$(COMPOSE_E2E) build frontend openmockllm-mistral $(cache)
 .PHONY: build-e2e
 
 down: ## stop and remove containers, networks, images, and volumes
@@ -138,7 +138,7 @@ logs: ## display app-dev logs (follow mode)
 .PHONY: logs
 
 run-backend: ## Start only the backend application and all needed services
-	@$(COMPOSE) up --force-recreate -d nginx ml-flow app-dev
+	@$(COMPOSE) up --force-recreate -d nginx app-dev
 .PHONY: run-backend
 
 run-frontend: ## Start only the frontend application
@@ -151,10 +151,14 @@ run:
 	@$(MAKE) run-frontend
 .PHONY: run
 
+create-compose-with-models: ## override the docker-compose file with models
+	cp -n compose.with_model.override.yml compose.override.yml
+.PHONY: create-compose-with-models
+
 run-e2e: ## start the e2e server
 run-e2e:
 	@$(MAKE) run-backend
-	@$(COMPOSE_E2E) up --force-recreate -d frontend
+	@$(COMPOSE_E2E) up --force-recreate -d frontend openmockllm-mistral
 .PHONY: run-e2e
 
 status: ## an alias for "docker compose ps"
@@ -227,7 +231,7 @@ superuser: ## Create an admin superuser with password "admin"
 .PHONY: superuser
 
 back-i18n-compile: ## compile the gettext files
-	@$(MANAGE) compilemessages --ignore="venv/**/*"
+	@$(MANAGE) compilemessages --ignore=".venv/**/*"
 .PHONY: back-i18n-compile
 
 back-i18n-generate: ## create the .pot files used for i18n

README.md

@@ -16,6 +16,24 @@
 </p>
 
 
+**Warning:** This project is in active development and in a very early stage. Breaking changes may occur at any time.
+
+
+## Yet another AI chatbot
+
+Conversations is an open-source AI chatbot designed to be simple, secure and privacy-friendly.
+
+Why another AI chatbot? Because we want to be able to fully control our data and the way we interact with AI.
+We want to have a very friendly end-user interface and code, and we want to be able to easily customize the
+chatbot to our needs.
+
+We leverage open-source projects such as [Vercel&lsquo;s AI SDK](https://ai-sdk.dev/) and [Pydantic AI](https://ai.pydantic.dev)
+and only assemble them in a way that makes sense for us and allows us to focus on the product.
+
+This assistant's purpose is also to be integrated into the "La Suite numérique" ecosystem of tools for public services.
+
+Any help to improve the project is very welcome!
+
 
 ### Self-host
 🚀 Conversations is easy to install on your own servers
@@ -30,9 +48,9 @@ In the works: Docker Compose, soon YunoHost
 
 You can test Conversations on your browser by visiting this => TBD
 
-### Run Docs locally
+### Run Conversations locally
 
-> ⚠️ The methods described below for running Docs locally is **for testing purposes only**. It is based on building Docs using [Minio](https://min.io/) as an S3-compatible storage solution. Of course you can choose any S3-compatible storage solution.
+> ⚠️ The methods described below for running Conversations locally is **for testing purposes only**.
 
 **Prerequisite**
 
@@ -97,6 +115,31 @@ To start all the services, except the frontend container, you can use the follow
 $ make run-backend
 ```
 
+**Setup a basic LLM call**
+
+To be able to use Conversations, you need to configure at least one Large Language Model (LLM) provider.
+You can do so by setting the appropriate environment variables in the `env.d/development/common` file:
+
+```ini
+AI_BASE_URL=http://host.docker.internal:12434/v1/
+AI_MODEL=gemma3:4b
+AI_API_KEY=XXX
+```
+
+for a local ollama, or by running a local LLM with docker-compose:
+
+```shellscript
+$ make create-compose-with-models
+```
+
+which will create a `compose.override.yml` file to start a local models `ai/smollm2`
+which can be changed later by editing the `compose.override.yml` file.
+
+You will need to call `make run` after changing the `env.d/development/common`
+or `compose.override.yml` file.
+
+You can find more information about configuring LLM providers in the [LLM Configuration](docs/llm-configuration.md) documentation.
+
 **Adding content**
 
 You can create a basic demo site by running this command:
@@ -123,6 +166,18 @@ You first need to create a superuser account:
 $ make superuser
 ```
 
+## Documentation 📚
+
+Additional documentation is available in the `docs/` directory:
+
+- [LLM Configuration](docs/llm-configuration.md) - Configure Large Language Models and providers
+- [Attachments](docs/attachments.md) - How to use attachments in conversations
+- [Tools for Agents](docs/tools.md) - Available tools and how to add new ones
+- [Environment Variables](docs/env.md) - All available environment variables
+- [Installation Guide](docs/installation.md) - Deploy on a Kubernetes cluster
+- [Theming](docs/theming.md) - Customize the application appearance
+- [Architecture](docs/architecture.md) - Technical architecture overview
+
 ## Licence 📝
 
 This work is released under the MIT License (see [LICENSE](https://github.com/suitenumerique/conversations/blob/main/LICENSE)).
@@ -152,9 +207,7 @@ docs
 
 ### Stack
 
-Conversations is built on top of [Django Rest Framework](https://www.django-rest-framework.org/), [Next.js](https://nextjs.org/), [Vercel&lsquo;s AI SDK](https://ai-sdk.dev/) and [OpenAI Agents SDK](https://github.com/openai/openai-agents-python). We thank the contributors of all these projects for their awesome work!
-
-
+Conversations is built on top of [Django Rest Framework](https://www.django-rest-framework.org/), [Next.js](https://nextjs.org/), [Vercel&lsquo;s AI SDK](https://ai-sdk.dev/) and [Pydantic AI](https://ai.pydantic.dev). We thank the contributors of all these projects for their awesome work!
 
 
 ### Gov ❤️ open source

SECURITY.md

@@ -6,7 +6,7 @@ Security is very important to us.
 
 If you have any issue regarding security, please disclose the information responsibly submitting [this form](https://vdp.numerique.gouv.fr/p/Send-a-report?lang=en) and not by creating an issue on the repository. You can also email us at docs@numerique.gouv.fr
 
-We appreciate your effort to make Docs more secure.
+We appreciate your effort to make Conversations more secure.
 
 ## Vulnerability disclosure policy
 
@@ -20,4 +20,4 @@ and Exposures (CVE) identifier for the vulnerability.
 3. Once this grace period has passed, we will publish the vulnerability.
 
 By adhering to this security policy, we aim to address security concerns
-effectively and responsibly in our open source software project.
+effectively and responsibly in our open source software project.

bin/Tiltfile

@@ -10,10 +10,6 @@ docker_build(
     target = 'backend-production',
     live_update=[
         sync('../src/backend', '/app'),
-        run(
-            'pip install -r /app/requirements.txt',
-            trigger=['./api/requirements.txt']
-        )
     ]
 )
 
@@ -28,9 +24,9 @@ docker_build(
     ]
 )
 
-k8s_resource('conversations-conversations-backend-migrate', resource_deps=['postgres-postgresql'])
-k8s_resource('conversations-conversations-backend-createsuperuser', resource_deps=['conversations-conversations-backend-migrate'])
-k8s_resource('conversations-conversations-backend', resource_deps=['conversations-conversations-backend-migrate'])
+k8s_resource('conversations-backend-migrate', resource_deps=['postgres-postgresql'])
+k8s_resource('conversations-backend-createsuperuser', resource_deps=['conversations-backend-migrate'])
+k8s_resource('conversations-backend', resource_deps=['conversations-backend-migrate'])
 k8s_yaml(local('cd ../src/helm && helmfile -n conversations -e dev template .'))
 
 migration = '''

compose-e2e.yml

@@ -11,3 +11,22 @@ services:
     image: conversations:frontend-production
     ports:
       - "3000:3000"
+
+  openmockllm-mistral:
+    user: "${DOCKER_USER:-1000}"
+    build:
+      context: .
+      dockerfile: ./src/OpenMockLLM/Dockerfile
+    image: conversations:openmockllm-mistral
+    command:
+      - openmockllm
+      - --host
+      - "0.0.0.0"
+      - --port
+      - "8000"
+      - --backend
+      - mistral
+      - --model-name
+      - mistral-mock
+    ports:
+      - "8900:8000"

compose.with_model.override.yml

@@ -0,0 +1,12 @@
+name: conversations
+
+services:
+  app-dev:
+    models:
+      llm:
+        endpoint_var: AI_BASE_URL
+        model_var: AI_MODEL
+
+models:
+  llm:
+    model: ai/smollm2

compose.yml

@@ -71,9 +71,13 @@ services:
       - "host.docker.internal:host-gateway"
     ports:
       - "8071:8000"
+    networks:
+      - default
+      - lasuite
     volumes:
       - ./src/backend:/app
       - ./data/static:/data/static
+      - /app/.venv
     depends_on:
         postgresql:
             condition: service_healthy
@@ -84,13 +88,14 @@ services:
           condition: service_started
         createbuckets:
           condition: service_started
-        ai_runner:
-          condition: service_started
 
   nginx:
     image: nginx:1.25
     ports:
       - "8083:8083"
+    networks:
+      - default
+      - lasuite
     volumes:
       - ./docker/files/etc/nginx/conf.d:/etc/nginx/conf.d:ro
     depends_on:
@@ -180,19 +185,7 @@ services:
         condition: service_healthy
         restart: true
 
-  ml-flow:
-    image: ghcr.io/mlflow/mlflow:v3.0.0
-    environment:
-      MLFLOW_TRACKING_URI: http://localhost:5050
-      MLFLOW_ARTIFACT_ROOT: /mlflow/artifacts
-    ports:
-      - "5050:5050"
-    volumes:
-      - ./data/mlflow:/mlflow/artifacts
-    command: mlflow server --host 0.0.0.0:5050
-
-  ai_runner:
-    provider:
-      type: model
-      options:
-        model: ai/smollm2
+networks:
+  lasuite:
+    name: lasuite-network
+    driver: bridge

docker/files/etc/nginx/conf.d/default.conf

@@ -25,7 +25,7 @@ server {
     }
 
     location /media-auth {
-        proxy_pass http://app-dev:8000/api/v1.0/documents/media-auth/;
+        proxy_pass http://app-dev:8000/api/v1.0/chats/media-auth/;
         proxy_set_header Host $host;
         proxy_set_header X-Real-IP $remote_addr;
         proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;

docs/architecture.md

@@ -6,14 +6,9 @@
 flowchart TD
     User -- HTTP --> Front("Frontend (NextJS SPA)")
     Front -- REST API --> Back("Backend (Django)")
-    Front -- WebSocket --> Yserver("Microservice Yjs (Express)") -- WebSocket -->  CollaborationServer("Collaboration server (Hocuspocus)") -- REST API <--> Back
     Front -- OIDC --> Back -- OIDC ---> OIDC("Keycloak / ProConnect")
-    Back -- REST API --> Yserver
     Back --> DB("Database (PostgreSQL)")
-    Back <--> Celery --> DB
+    Back --> Cache("Cache (Redis)")
     Back ----> S3("Minio (S3)")
+    Back -- REST API --> LLM("LLM Providers")
 ```
-
-### Architecture decision records
-
-- [ADR-0001-20250106-use-yjs-for-docs-editing](./adr/ADR-0001-20250106-use-yjs-for-docs-editing.md)

docs/attachments.md

@@ -0,0 +1,400 @@
+# Conversation Attachments
+
+This document describes how conversation attachments work in the Conversations application, including the upload process, security measures, and how documents are processed for use with Large Language Models (LLMs).
+
+## Table of Contents
+
+- [Overview](#overview)
+- [Supported Attachment Types](#supported-attachment-types)
+- [Architecture & Flow](#architecture--flow)
+  - [High-Level Overview](#high-level-overview)
+  - [Detailed Technical Flow](#detailed-technical-flow)
+- [Security & Validation](#security--validation)
+  - [MIME Type Validation](#mime-type-validation)
+  - [Malware Detection](#malware-detection)
+- [Document Processing for LLMs](#document-processing-for-llms)
+  - [Image Attachments](#image-attachments)
+  - [PDF Documents](#pdf-documents)
+  - [Other Document Types](#other-document-types)
+- [Configuration](#configuration)
+
+---
+
+## Overview
+
+Conversations allows users to attach files to their conversations with the AI assistant. These attachments can be:
+- **Images** (displayed directly to vision-capable LLMs)
+- **PDF documents** (sent as document URLs to the LLM)
+- **Other documents** (converted to text and indexed for semantic search)
+
+The attachment system uses **S3-compatible object storage** (such as MinIO in development) to store files securely. 
+The backend generates **presigned URLs** that allow the frontend to upload files directly to the storage, 
+without routing the file data through the backend server.
+
+Note about documents: The system uses a tool called **MarkItDown** to convert various document formats 
+(Word, Excel, PowerPoint, text files, etc.) into Markdown text for processing by LLMs. When at least 
+one non-PDF/image document is attached, the system enables:
+ - a **Retrieval-Augmented Generation (RAG)** search tool to allow the LLM to query relevant sections of the documents.
+ - a **summarization tool** to provide document summaries on user request.
+   ⚠️ naive implementation at the moment, needs improvement before being used in production.
+
+## Supported Attachment Types
+The following attachment types are supported:
+- **Images**: `image/png`, `image/jpeg`, `image/gif`, `image/webp`.
+- **PDF documents**: `application/pdf`
+- **Other documents**:
+  - Microsoft Word: `application/vnd.openxmlformats-officedocument.wordprocessingml.document`
+  - Microsoft Excel: `application/vnd.openxmlformats-officedocument.spreadsheetml.sheet`
+  - Microsoft PowerPoint: `application/vnd.openxmlformats-officedocument.presentationml.presentation`
+  - Text files: `text/plain`, `text/markdown`, `text/csv`
+
+**Warning**: The current implementation for PDF expects the LLM to be able to manage them. We need to
+improve the handling of PDFs in case the LLM cannot process them natively.
+
+**Todo**: 
+ - Add support for more file types and improve document processing workflows.
+ - Allow PDF management via RAG search when the LLM cannot handle them natively.
+ - Allow file type restrictions based on model settings, instead of globally.
+ - Improve the summarization tool to provide better summaries and handle larger documents.
+ - Start file upload right away when the user selects a file, instead of waiting for the user to send the message.
+
+
+---
+
+## Architecture & Flow
+
+### High-Level Overview
+
+```
+┌─────────────┐       ┌─────────────┐       ┌─────────────┐       ┌─────────────┐
+│   Frontend  │       │   Backend   │       │  S3 Storage │       │ Malware Det.│
+└──────┬──────┘       └──────┬──────┘       └──────┬──────┘       └──────┬──────┘
+       │                     │                     │                     │
+       │ 1. Create attachment│                     │                     │
+       ├────────────────────>│                     │                     │
+       │                     │                     │                     │
+       │ 2. Return presigned │                     │                     │
+       │    URL for upload   │                     │                     │
+       │<────────────────────┤                     │                     │
+       │                     │                     │                     │
+       │ 3. Upload file      │                     │                     │
+       │    directly to S3   │                     │                     │
+       ├──────────────────────────────────────────>│                     │
+       │                     │                     │                     │
+       │ 4. Notify upload    │                     │                     │
+       │    completed        │                     │                     │
+       ├────────────────────>│                     │                     │
+       │                     │                     │                     │
+       │                     │ 5. Detect MIME type │                     │
+       │                     ├────────────────────>│                     │
+       │                     │                     │                     │
+       │                     │ 6. Scan for malware │                     │
+       │                     ├──────────────────────────────────────────>│
+       │                     │                     │                     │
+       │                     │ 7. Update status    │                     │
+       │ 8. Return status    │<──────────────────────────────────────────┤
+       │<────────────────────┤                     │                     │
+       │                     │                     │                     │
+```
+
+### Detailed Technical Flow
+
+#### Step 1: Attachment Creation Request
+
+When a user selects a file to upload, the frontend sends a POST request to create an attachment record:
+
+**Endpoint**: `POST /api/conversations/{conversation_id}/attachments/`
+
+**Request payload**:
+```json
+{
+  "file_name": "document.pdf",
+  "size": 1048576,
+  "content_type": "application/pdf"
+}
+```
+
+**Backend processing** (`ChatConversationAttachmentViewSet.perform_create`):
+1. Verifies the user owns the conversation
+2. Generates a unique UUID for the file
+3. Creates a storage key: `{conversation_id}/attachments/{uuid}.{extension}`
+4. Creates a database record with status `PENDING`
+
+**Response**:
+```json
+{
+  "id": "uuid-of-attachment",
+  "key": "conversation-id/attachments/file-id.pdf",
+  "file_name": "document.pdf",
+  "size": 1048576,
+  "upload_state": "pending",
+  "policy": "https://s3.example.com/bucket/...?presigned-params"
+}
+```
+
+The `policy` field contains a **presigned URL** valid for a limited time (configured by `AWS_S3_UPLOAD_POLICY_EXPIRATION`).
+
+#### Step 2: Direct Upload to S3
+
+The frontend uses the presigned URL to upload the file directly to S3 storage using a PUT request.
+
+**Technical details**:
+- The presigned URL includes authentication parameters
+- The upload is done with `Content-Type` header matching the file's MIME type
+- No backend involvement in the data transfer
+
+#### Step 3: Upload Completion Notification
+
+After successful upload, the frontend notifies the backend:
+
+**Endpoint**: `POST /api/conversations/{conversation_id}/attachments/{attachment_id}/upload-ended/`
+
+**Backend processing** (`ChatConversationAttachmentViewSet.upload_ended`):
+
+1. **MIME Type Detection** (`chat/views.py`):
+   ```python
+   mime_detector = magic.Magic(mime=True)
+   with default_storage.open(attachment.key, "rb") as file:
+       mimetype = mime_detector.from_buffer(file.read(2048))
+       size = file.size
+   ```
+   
+   Uses `python-magic` to detect the actual MIME type from file content (first 2048 bytes).
+
+2. **Update attachment status**:
+   - Status: `PENDING` → `ANALYZING`
+   - Store detected MIME type and actual file size
+
+3. **Trigger Malware Detection**:
+   ```python
+   malware_detection.analyse_file(
+       attachment.key,
+       safe_callback="chat.malware_detection.conversation_safe_attachment_callback",
+       unknown_callback="chat.malware_detection.unknown_attachment_callback",
+       unsafe_callback="chat.malware_detection.conversation_unsafe_attachment_callback",
+       conversation_id=conversation_id,
+   )
+   ```
+
+#### Step 4: Malware Detection Callbacks
+
+The malware detection service (configurable via `MALWARE_DETECTION_BACKEND`) scans the file and calls one of three callbacks:
+
+**Safe file** (`conversation_safe_attachment_callback`):
+- Status: `ANALYZING` → `READY`
+- File is ready for use
+
+**Unsafe file** (`conversation_unsafe_attachment_callback`):
+- Status: `ANALYZING` → `SUSPICIOUS`
+- File is quarantined and not accessible
+- Security log entry created
+
+**Unknown status** (`unknown_attachment_callback`):
+- Handles special cases (e.g., file too large to analyze)
+- Status: `ANALYZING` → `FILE_TOO_LARGE_TO_ANALYZE`
+
+---
+
+## Security & Validation
+
+For now, the system is not intended to host user-uploaded files for public download.
+All files are stored in private S3 buckets with presigned URLs for controlled access and only
+the owner of the conversation/the uploader can access them, so the risk is quite low around bad use of
+the attachment system.
+
+Also, the document content is sent to the LLM and does not prevent any prompt injection attacks, which is not
+an issue specific to the attachment system but to the overall design of LLM-based applications and should be
+addressed globally. Also for the moment, the system does not have any action tools that could be used to execute
+malicious code based on document content.
+
+### Malware Detection
+
+The malware detection system is **pluggable** and configurable, allowing different backends to be used.
+By default, a `DummyBackend` is provided that marks all files as safe.
+
+⚠️ The current implementation does not disallow any file types or status from being used in conversations.
+This is a potential security risk and should be addressed in future versions.
+
+---
+
+## Document Processing for LLMs
+
+When a user sends a message with attachments, the system processes them differently based on their type:
+
+### Image Attachments
+
+**MIME types**: `image/png`, `image/jpeg`, `image/gif`, `image/webp`, etc.
+
+**Processing flow**:
+
+1. **URL Conversion**: Local media URLs are converted to presigned S3 URLs before sending to the LLM:
+   ```python
+   # From: chat/agents/local_media_url_processors.py
+   content.url = generate_retrieve_policy(key)
+   ```
+
+2. **Sent to LLM**: Images are sent as `ImageUrl` objects in the prompt:
+   ```python
+   ImageUrl(
+       url="https://s3.example.com/bucket/key?presigned-params",
+       identifier="file-id.png",
+   )
+   ```
+
+3. **Vision models** can analyze the image content directly.
+
+4. **Response processing**: After the LLM responds, presigned URLs are converted back to local URLs for storage:
+   ```python
+   # Mapping: presigned_url -> /media-key/{conversation_id}/attachments/{file_id}.png
+   ```
+
+### PDF Documents
+
+**MIME type**: `application/pdf`
+
+**Processing flow**:
+
+1. **Direct URL passing**: PDFs are sent as `DocumentUrl` objects :
+   ```python
+   DocumentUrl(
+       url="https://s3.example.com/bucket/key?presigned-params",
+       identifier="file-id.pdf",
+   )
+   ```
+
+2. **LLM processing**: Compatible LLMs can:
+   - Extract and read text from PDFs
+   - Understand document structure
+   - Answer questions about the content
+
+3. **No conversion needed**: PDFs are passed directly without preprocessing.
+
+### Other Document Types
+
+**MIME types**: Word documents, Excel spreadsheets, PowerPoint, text files, Markdown, etc.
+
+**Processing flow**:
+
+1. **Document parsing**: When a document is uploaded, it's parsed using the `AlbertRagBackend` class.
+
+2. **Conversion to Markdown**: Documents are converted using **MarkItDown** library or using the "Albert API" for PDFs.
+
+3. **RAG (Retrieval-Augmented Generation)**:
+   - Converted text is indexed in a vector database
+   - The LLM uses a `document_rag_search` tool to query relevant sections
+   - Only relevant chunks are sent to the LLM to fit context windows
+
+4. **Summarization tool** if needed.
+
+### Processing Strategy Decision Tree
+
+**Decision logic**:
+- **No documents**: Standard conversation
+- **Images**: Send as direct (presigned) URLs to the LLM
+- **Only PDFs**: Send as direct (presigned) URLs to the LLM
+- **Other documents present**: Enable RAG search tool + convert to Markdown
+
+---
+
+## Configuration
+
+### Environment Variables
+
+| Variable                                     | Default        | Description                                                |
+|----------------------------------------------|----------------|------------------------------------------------------------|
+| `ATTACHMENT_MAX_SIZE`                        | Configurable   | Maximum file size in bytes                                 |
+| `ATTACHMENT_CHECK_UNSAFE_MIME_TYPES_ENABLED` | `True`         | Enable/disable MIME type validation                        |
+| `AWS_S3_UPLOAD_POLICY_EXPIRATION`            | 3600           | Presigned URL expiration (seconds)                         |
+| `AWS_S3_RETRIEVE_POLICY_EXPIRATION`          | 3600           | Presigned retrieval URL expiration (seconds)               |
+| `AWS_S3_DOMAIN_REPLACE`                      | None           | Alternative S3 domain for presigned URLs (for development) |
+| `MALWARE_DETECTION_BACKEND`                  | `DummyBackend` | Malware scanning backend class                             |
+| `MALWARE_DETECTION_PARAMETERS`               | `{}`           | Backend-specific configuration                             |
+| `RAG_FILES_ACCEPTED_FORMATS`                 | See below      | List of MIME types accepted for file uploads               |
+
+#### RAG_FILES_ACCEPTED_FORMATS
+
+This environment variable controls which file types users are allowed to upload as attachments to conversations.
+
+**Configuration**:
+- **Type**: List of strings (comma-separated MIME types when using environment variable)
+- **Default value**: Includes a comprehensive list of document and image formats:
+  - Microsoft Office documents (`.docx`, `.pptx`, `.xlsx`, `.xls`)
+  - Text files (`.txt`, `.csv`)
+  - PDF documents (`.pdf`)
+  - HTML files
+  - Markdown files (`.md`)
+  - Outlook messages (`.msg`)
+  - Images (`.jpeg`, `.png`, `.gif`, `.webp`)
+
+**Example configuration**:
+```ini
+# In environment variable (comma-separated)
+RAG_FILES_ACCEPTED_FORMATS="application/pdf,text/plain,image/png,image/jpeg"
+```
+
+```python
+# In Django settings (as a Python list)
+RAG_FILES_ACCEPTED_FORMATS = [
+    "application/pdf",
+    "text/plain",
+    "image/png",
+    "image/jpeg",
+]
+```
+
+**How it's used**:
+1. **Backend**: The list is exposed via the `/api/v1.0/config/` endpoint as `chat_upload_accept` (MIME types joined with commas)
+2. **Frontend**: The configuration is used to validate files before upload in the chat interface:
+   - Checks exact MIME type matches
+   - Supports wildcard patterns (e.g., `image/*` for all image types)
+   - Supports file extension patterns (e.g., `.pdf`)
+3. **User experience**: Files that don't match the accepted formats are rejected with a user-friendly error message
+
+**Notes**:
+
+ - This setting controls frontend validation only. Backend validation should also be implemented for security.
+ - Future improvements may include per-model file type restrictions.
+
+### Storage Configuration
+
+**MinIO (Development)**:
+```yaml
+# docker-compose.yml
+minio:
+  image: minio/minio
+  environment:
+    MINIO_ROOT_USER: minioadmin
+    MINIO_ROOT_PASSWORD: minioadmin
+  command: server /data --console-address ":9001"
+```
+
+---
+
+## Troubleshooting
+
+### LLM Cannot Access Image/PDF
+
+**Possible causes**:
+- Presigned URL has expired
+- S3 storage is not accessible from the LLM provider
+- CORS configuration issues
+
+**Solution**: Check `AWS_S3_RETRIEVE_POLICY_EXPIRATION` and S3 access policies.
+
+### Document Not Appearing in RAG Search
+
+**Possible causes**:
+- Document conversion failed
+- Vector database indexing failed
+
+**Check logs**: Look for errors in `DocumentConverter` and RAG backend logs.
+
+---
+
+## Related Documentation
+
+- [Installation Guide](installation.md) - S3 storage setup
+- [LLM Configuration](llm-configuration.md) - Model capabilities for attachments
+- [Architecture](architecture.md) - System overview
+- [Tools](tools.md) - Document search and RAG tools
+

docs/env.md

@@ -10,7 +10,6 @@ These are the environment variables you can set for the `conversations-backend`
 |-------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------|---------------------------------------------------------|
 | DJANGO_ALLOWED_HOSTS                            | allowed hosts                                                                                                                     | []                                                      |
 | DJANGO_SECRET_KEY                               | secret key                                                                                                                        |                                                         |
-| DJANGO_SERVER_TO_SERVER_API_TOKENS              |                                                                                                                                   | []                                                      |
 | DB_ENGINE                                       | engine to use for database connections                                                                                            | django.db.backends.postgresql_psycopg2                  |
 | DB_NAME                                         | name of the database                                                                                                              | conversations                                           |
 | DB_USER                                         | user to authenticate with                                                                                                         | dinum                                                   |
@@ -24,12 +23,11 @@ These are the environment variables you can set for the `conversations-backend`
 | AWS_S3_SECRET_ACCESS_KEY                        | access key for s3 endpoint                                                                                                        |                                                         |
 | AWS_S3_REGION_NAME                              | region name for s3 endpoint                                                                                                       |                                                         |
 | AWS_STORAGE_BUCKET_NAME                         | bucket name for s3 endpoint                                                                                                       | conversations-media-storage                             |
-| DOCUMENT_IMAGE_MAX_SIZE                         | maximum size of document in bytes                                                                                                 | 10485760                                                |
+| ATTACHMENT_MAX_SIZE                             | maximum size of document in bytes                                                                                                 | 10485760                                                |
 | LANGUAGE_CODE                                   | default language                                                                                                                  | en-us                                                   |
 | API_USERS_LIST_THROTTLE_RATE_SUSTAINED          | throttle rate for api                                                                                                             | 180/hour                                                |
 | API_USERS_LIST_THROTTLE_RATE_BURST              | throttle rate for api on burst                                                                                                    | 30/minute                                               |
 | SPECTACULAR_SETTINGS_ENABLE_DJANGO_DEPLOY_CHECK |                                                                                                                                   | false                                                   |
-| TRASHBIN_CUTOFF_DAYS                            | trashbin cutoff                                                                                                                   | 30                                                      |
 | DJANGO_EMAIL_BACKEND                            | email backend library                                                                                                             | django.core.mail.backends.smtp.EmailBackend             |
 | DJANGO_EMAIL_BRAND_NAME                         | brand name for email                                                                                                              |                                                         |
 | DJANGO_EMAIL_HOST                               | host name of email                                                                                                                |                                                         |
@@ -76,13 +74,14 @@ These are the environment variables you can set for the `conversations-backend`
 | OIDC_USERINFO_FULLNAME_FIELDS                   | OIDC token claims to create full name                                                                                             | ["first_name", "last_name"]                             |
 | OIDC_USERINFO_SHORTNAME_FIELD                   | OIDC token claims to create shortname                                                                                             | first_name                                              |
 | ALLOW_LOGOUT_GET_METHOD                         | Allow get logout method                                                                                                           | true                                                    |
-| AI_API_KEY                                      | AI key to be used for AI Base url                                                                                                 |                                                         |
-| AI_BASE_URL                                     | OpenAI compatible AI base url                                                                                                     |                                                         |
-| AI_MODEL                                        | AI Model to use                                                                                                                   |                                                         |
-| AI_AGENT_NAME                                   | Name of the AI agent (useless)                                                                                                    | Conversations Assistant                                 |
-| AI_AGENT_INSTRUCTION                            | Base instruction for the AI agent                                                                                                 | You are a helpful assistant                             |
-| Y_PROVIDER_API_KEY                              | Y provider API key                                                                                                                |                                                         |
-| Y_PROVIDER_API_BASE_URL                         | Y Provider url                                                                                                                    |                                                         |
+| LLM_CONFIGURATION_FILE_PATH                     | Path to the LLM configuration JSON file. See [LLM Configuration](llm-configuration.md) for details                                | <BASE_DIR>/conversations/configuration/llm/default.json |
+| LLM_DEFAULT_MODEL_HRID                          | HRID of the model used for conversations                                                                                          | default-model                                           |
+| LLM_SUMMARIZATION_MODEL_HRID                    | HRID of the model used for summarization                                                                                          | default-summarization-model                             |
+| AI_API_KEY                                      | AI API key to be used for the default provider (used in default LLM configuration, not for production use)                        |                                                         |
+| AI_BASE_URL                                     | OpenAI compatible AI base URL (used in default LLM configuration, not for production use)                                         |                                                         |
+| AI_MODEL                                        | AI Model name to use (used in default LLM configuration, not for production use)                                                  |                                                         |
+| AI_AGENT_INSTRUCTIONS                           | Base instruction for the AI agent (used in default LLM configuration, not for production use)                                     | You are a helpful assistant. Wrap formulas...           |
+| AI_AGENT_TOOLS                                  | List of enabled tools for the agent (used in default LLM configuration, not for production use)                                   | []                                                      |
 | CONVERSION_API_ENDPOINT                         | Conversion API endpoint                                                                                                           | convert-markdown                                        |
 | CONVERSION_API_CONTENT_FIELD                    | Conversion api content field                                                                                                      | content                                                 |
 | CONVERSION_API_TIMEOUT                          | Conversion api timeout                                                                                                            | 30                                                      |
@@ -96,6 +95,9 @@ These are the environment variables you can set for the `conversations-backend`
 | CACHES_KEY_PREFIX                               | The prefix used to every cache keys.                                                                                              | conversations                                           |
 | THEME_CUSTOMIZATION_FILE_PATH                   | full path to the file customizing the theme. An example is provided in src/backend/conversations/configuration/theme/default.json | BASE_DIR/conversations/configuration/theme/default.json |
 | THEME_CUSTOMIZATION_CACHE_TIMEOUT               | Cache duration for the customization settings                                                                                     | 86400                                                   |
+| FIND_API_KEY                                    | API key of Find                                                                                                                   |                                                         |
+| FIND_API_URL                                    | URL of Find                                                                                                                       | `https://app-find/api`                                  |
+| FIND_API_TIMEOUT                                | Find API timeout                                                                                                                  | 30                                                      |
 
 
 ## conversations-frontend image

docs/examples/conversations.values.yaml

@@ -9,7 +9,6 @@ backend:
     DJANGO_CSRF_TRUSTED_ORIGINS: https://conversations.127.0.0.1.nip.io
     DJANGO_CONFIGURATION: Feature
     DJANGO_ALLOWED_HOSTS: conversations.127.0.0.1.nip.io
-    DJANGO_SERVER_TO_SERVER_API_TOKENS: secret-api-key
     DJANGO_SECRET_KEY: AgoodOrAbadKey
     DJANGO_SETTINGS_MODULE: conversations.settings
     DJANGO_SUPERUSER_PASSWORD: admin
@@ -123,7 +122,7 @@ ingressMedia:
   host: conversations.127.0.0.1.nip.io
 
   annotations:
-    nginx.ingress.kubernetes.io/auth-url: https://conversations.127.0.0.1.nip.io/api/v1.0/documents/media-auth/
+    nginx.ingress.kubernetes.io/auth-url: https://conversations.127.0.0.1.nip.io/api/v1.0/chats/media-auth/
     nginx.ingress.kubernetes.io/auth-response-headers: "Authorization, X-Amz-Date, X-Amz-Content-SHA256"
     nginx.ingress.kubernetes.io/upstream-vhost: minio.conversations.svc.cluster.local:9000
     nginx.ingress.kubernetes.io/rewrite-target: /conversations-media-storage/$1

docs/file-upload-modes.md

@@ -0,0 +1,159 @@
+# File Upload Modes
+
+This document describes the different modes for handling file uploads in the Conversations application, and how to configure and use them.
+
+## Overview
+
+The application supports two independent configuration points:
+
+1. **`FILE_UPLOAD_MODE`**: how the frontend uploads files (frontend → storage/backend)
+2. **`FILE_TO_LLM_MODE`**: how the backend provides files to the LLM (backend → LLM)
+
+Each mode has different trade-offs in terms of security, performance, and LLM accessibility. The two settings can be combined based on your network constraints.
+
+## Configuration
+
+### Frontend upload mode (`FILE_UPLOAD_MODE`)
+
+```bash
+# Default: presigned URL upload (backward compatible)
+FILE_UPLOAD_MODE=presigned_url
+
+# Frontend uploads directly to backend
+FILE_UPLOAD_MODE=backend_to_s3
+```
+
+### Backend delivery mode (`FILE_TO_LLM_MODE`)
+
+```bash
+# Default: presigned URL mode (backward compatible)
+FILE_TO_LLM_MODE=presigned_url
+
+# Backend provides base64-encoded data URLs
+FILE_TO_LLM_MODE=backend_base64
+
+# Backend provides temporary URLs through the backend
+FILE_TO_LLM_MODE=backend_temporary_url
+```
+
+Additional settings for backend temporary URL mode:
+
+```bash
+# Base URL to reach backend
+FILE_BACKEND_URL="http://localhost:8071"
+
+# Expiration time for temporary URLs (in seconds, default: 180 = 3 minutes)
+FILE_BACKEND_TEMPORARY_URL_EXPIRATION=180
+```
+
+## Mode Details
+
+### 1. Presigned URL Mode (Default)
+
+**Frontend upload configuration:** `FILE_UPLOAD_MODE=presigned_url`
+
+**Backend delivery configuration:** `FILE_TO_LLM_MODE=presigned_url`
+
+**How it works:**
+- Frontend requests a presigned URL from the backend
+- Frontend uploads the file directly to S3 using the presigned URL
+- Frontend notifies the backend when upload is complete
+- Backend initiates malware detection
+- Backend returns presigned S3 URLs to the LLM
+
+**Advantages:**
+- Files don't pass through the backend server (lower bandwidth usage)
+- Faster uploads for large files (direct to S3)
+- S3 handles the upload, no backend load
+- Backward compatible with existing frontend implementations
+
+**Disadvantages:**
+- S3 bucket must be accessible from the frontend
+- Presigned URLs can be leaked if not handled carefully
+- Frontend needs to handle S3 credentials/configuration
+
+**LLM Access:**
+- Images: Presigned S3 URLs with expiration (default: 3 minutes)
+- Documents: Presigned S3 URLs with expiration (default: 3 minutes)
+
+**When to use:**
+- When frontend has direct access to S3
+- When you want to minimize backend load
+- When S3 is publicly accessible or accessible via VPN
+
+
+### 2. Backend Base64 Mode
+
+**Frontend upload configuration:** `FILE_UPLOAD_MODE=backend_to_s3`
+
+**Backend delivery configuration:** `FILE_TO_LLM_MODE=backend_base64`
+
+**How it works:**
+- Frontend uploads the file directly to the backend
+- Backend stores the file on S3
+- Backend reads the file, encodes it as base64, and creates a data URL
+- LLM receives the file as a base64-encoded data URL
+
+**Advantages:**
+- S3 can be private/internal (not accessible from frontend)
+- Files always go through the backend for validation
+- No presigned URLs to manage
+- Better control over file access
+- Data URLs work with all LLMs that support file content
+
+**Disadvantages:**
+- Backend memory usage increases (entire file loaded for base64 encoding)
+- Slower for very large files (encoding overhead)
+- Increased bandwidth on backend
+- Data URLs can be very large in responses
+
+**LLM Access:**
+- Images: Base64-encoded data URLs (format: `data:image/png;base64,...`)
+- Documents: Base64-encoded data URLs (format: `data:application/pdf;base64,...`)
+
+**When to use:**
+- When S3 is not accessible from the frontend
+- When you want all file uploads to go through the backend
+- When the LLM supports base64-encoded data URLs
+- For smaller files (< 50MB)
+
+
+### 3. Backend Temporary URL Mode
+
+**Frontend upload configuration:** `FILE_UPLOAD_MODE=backend_to_s3`
+
+**Backend delivery configuration:** `FILE_TO_LLM_MODE=backend_temporary_url`
+
+**How it works:**
+- Frontend uploads the file directly to the backend
+- Backend stores the file on S3
+- Backend generates a secure temporary access token stored in cache (TTL: 3 minutes by default)
+- Backend returns a temporary URL pointing to the backend's file-stream endpoint
+- LLM receives the temporary URL and accesses the file through the backend
+- Backend validates the token and streams the file content from S3 to the LLM
+
+**Advantages:**
+- S3 can be private/internal (not accessible from frontend or LLM directly)
+- Files always go through the backend for validation and access control
+- LLM doesn't need direct access to S3
+- Tokens expire quickly (better security than long-lived presigned URLs)
+- No large data URL strings in memory or responses
+- Lower backend memory usage than base64 mode
+- Centralized file access control through the backend
+- Good balance between security and performance
+
+**Disadvantages:**
+- LLM must be able to access the backend server
+- File streaming goes through the backend (adds some latency)
+- Time-limited access (token expires)
+
+**LLM Access:**
+- Images: Temporary backend URLs with format `/api/v1.0/file-stream/{temporary_key}/` (token expiration: configurable, default: 3 minutes)
+- Documents: Temporary backend URLs with format `/api/v1.0/file-stream/{temporary_key}/` (token expiration: configurable, default: 3 minutes)
+
+**When to use:**
+- When S3 is not accessible from the frontend or LLM
+- When you want backend control over uploads and file access
+- When you want time-limited access to files with centralized control
+- When you want the LLM to access files through the backend gateway
+- For large files (backend streams directly from S3 without loading entirely into memory)

docs/installation.md

@@ -7,7 +7,7 @@ This document is a step-by-step guide that describes how to install Conversation
 - k8s cluster with an nginx-ingress controller
 - an OIDC provider (if you don't have one, we provide an example)
 - a PostgreSQL server (if you don't have one, we provide an example)
-- a Memcached server (if you don't have one, we provide an example)
+- a Redis server (if you don't have one, we provide an example)
 - a S3 bucket (if you don't have one, we provide an example)
 
 ### Test cluster

docs/llm-configuration.md

@@ -0,0 +1,412 @@
+# LLM Configuration
+
+This document describes how to configure Large Language Models (LLMs) in Conversations via the configuration file.
+
+## Overview
+
+Conversations uses a JSON configuration file to define LLM models and providers. This approach allows you to:
+- Configure multiple LLM models from different providers
+- Switch between models without code changes
+- Customize model-specific settings like temperature, max tokens, and system prompts
+- Enable or disable models dynamically
+
+The overall structure consists of two main sections: `providers` and `models`.
+Settings for models, provides customization through `settings` and `profile`, which corresponds to the 
+Pydantic AI model settings and profile. While we currently not use those settings extensively, 
+they are available for future use and advanced configurations, please reach us if you face any problem using them.
+
+## Configuration File Location
+
+The default LLM configuration file is located at:
+```
+src/backend/conversations/configuration/llm/default.json
+```
+
+You can override this location by setting the `LLM_CONFIGURATION_FILE_PATH` environment variable, but be careful as 
+this path must be accessible by the backend application _inside the docker image_:
+``` ini
+LLM_CONFIGURATION_FILE_PATH=/path/to/your/llm/config.json
+```
+
+## Default Behavior
+
+### Default Configuration
+
+The default configuration file is useful for local development and running the test, while it can be used
+in production, we suggest to create a specific one for production and replace the `settings.` values with
+`environ.` one.
+
+The default configuration file (`default.json`) includes:
+
+1. **Two default models**:
+   - `default-model`: The primary conversational model used for chat interactions
+   - `default-summarization-model`: A specialized model for summarizing conversations
+
+2. **One default provider**:
+   - `default-provider`: An OpenAI-compatible provider that uses environment variables for configuration
+
+### Environment Variable Integration
+
+The configuration uses dynamic value resolution with two special prefixes:
+
+- `settings.VARIABLE_NAME`: Resolves to a Django setting value
+- `environ.VARIABLE_NAME`: Resolves to an environment variable value
+
+For example, in the default configuration:
+```json
+{
+  "model_name": "settings.AI_MODEL",
+  "system_prompt": "settings.AI_AGENT_INSTRUCTIONS",
+  "tools": "settings.AI_AGENT_TOOLS"
+}
+```
+
+This allows to configure models in tests using the setting override mechanism from Django/Pytest (but might be replaced
+later with a simple override of the full configuration like it's done in some tests already).
+
+### Required Environment Variables
+
+For the default configuration to work, you need to set these environment variables:
+
+| Variable                      | Description                            | Example                     |
+|-------------------------------|----------------------------------------|-----------------------------|
+| `AI_API_KEY`                  | API key for the default provider       | `sk-...`                    |
+| `AI_BASE_URL`                 | Base URL for the OpenAI-compatible API | `https://api.openai.com/v1` |
+| `AI_MODEL`                    | Model name to use                      | `gpt-4o-mini`               |
+
+### Optional Environment Variables
+
+If you want to customize the agent behavior and tools, you can set these optional environment variables 
+(defaults are provided in the default configuration):
+
+| Variable                      | Description                            | Default           |
+|-------------------------------|----------------------------------------|-------------------|
+| `AI_AGENT_INSTRUCTIONS`       | System prompt for the agent            | see `settings.py` |
+| `AI_AGENT_TOOLS`              | List of enabled tools                  | `[]`              |
+| `SUMMARIZATION_SYSTEM_PROMPT` | Base prompt of the summarization agent | see `settings.py` |
+
+### Model Selection
+
+You can configure which models are used for specific tasks via environment variables:
+
+| Variable                       | Description                              | Default                       |
+|--------------------------------|------------------------------------------|-------------------------------|
+| `LLM_DEFAULT_MODEL_HRID`       | HRID of the model used for conversations | `default-model`               |
+| `LLM_SUMMARIZATION_MODEL_HRID` | HRID of the model used for summarization | `default-summarization-model` |
+
+## Configuration Structure
+
+The configuration file has two main sections:
+
+### 1. Providers
+
+Providers define the API endpoints and authentication for LLM services.
+
+```json
+{
+  "providers": [
+    {
+      "hrid": "unique-provider-id",
+      "base_url": "https://api.example.com/v1",
+      "api_key": "environ.API_KEY_VAR",
+      "kind": "openai"
+    }
+  ]
+}
+```
+
+**Provider Fields:**
+
+| Field      | Type   | Required | Description                                             |
+|------------|--------|----------|---------------------------------------------------------|
+| `hrid`     | string | Yes      | Unique identifier for the provider                      |
+| `base_url` | string | Yes      | API base URL (can use `settings.` or `environ.` prefix) |
+| `api_key`  | string | Yes      | API authentication key (use `environ.` here)            |
+| `kind`     | string | Yes      | Provider type: `openai` or `mistral`                    |
+
+### 2. Models
+
+Models define the LLMs available in your application.
+
+```json
+{
+  "models": [
+    {
+      "hrid": "unique-model-id",
+      "model_name": "gpt-4o-mini",
+      "human_readable_name": "GPT-4o Mini",
+      "provider_name": "unique-provider-id",
+      "profile": null,
+      "settings": {},
+      "is_active": true,
+      "icon": null,
+      "system_prompt": "You are a helpful assistant",
+      "tools": []
+    }
+  ]
+}
+```
+
+**Model Fields:**
+
+| Field                 | Type         | Required | Description                                                                                         |
+|-----------------------|--------------|----------|-----------------------------------------------------------------------------------------------------|
+| `hrid`                | string       | Yes      | Unique identifier for the model                                                                     |
+| `model_name`          | string       | Yes      | Name of the model as recognized by the provider (can use `settings.` or `environ.` prefix)          |
+| `human_readable_name` | string       | Yes      | Display name shown to users                                                                         |
+| `provider_name`       | string       | No*      | Reference to a provider's `hrid`                                                                    |
+| `provider`            | object       | No*      | Inline provider definition (alternative to `provider_name`)                                         |
+| `profile`             | object       | No       | Model-specific capabilities and settings                                                            |
+| `settings`            | object       | No       | Model inference settings (temperature, max_tokens, etc.)                                            |
+| `is_active`           | boolean      | Yes      | Whether the model is available for use                                                              |
+| `icon`                | string/array | No       | Base64-encoded icon or array of icon parts                                                          |
+| `system_prompt`       | string       | Yes      | Default system prompt for the model (can use `settings.` or `environ.` prefix)                      |
+| `tools`               | array        | Yes      | List of enabled tools for this model (can use `settings.` or `environ.` prefix for the whole array) |
+| `supports_streaming`  | boolean      | No       | Whether the model supports streaming responses                                                      |
+
+\* Either `provider_name` or `provider` must be set, unless `model_name` is in the format `<provider>:<model>`.
+
+## Adding New Models
+
+### Example 1: Adding a New OpenAI Model
+
+To add a new OpenAI model using the existing default provider:
+
+```json
+{
+  "models": [
+    // ...existing models...
+    {
+      "hrid": "gpt-4-turbo",
+      "model_name": "gpt-4-turbo-preview",
+      "human_readable_name": "GPT-4 Turbo",
+      "provider_name": "default-provider",
+      "profile": null,
+      "settings": {
+        "temperature": 0.7,
+        "max_tokens": 4096
+      },
+      "is_active": true,
+      "icon": null,
+      "system_prompt": "You are an expert AI assistant.",
+      "tools": ["web_search_brave_with_document_backend"],
+      "supports_streaming": true
+    }
+  ],
+  "providers": [
+    // ...existing providers...
+  ]
+}
+```
+
+### Example 2: Adding a Model using Pydantic AI format
+
+To add a model with a specific provider using the default Pydantic AI format, you don't need to define the provider separately if you use the `model_name` format `<provider>:<model>`.
+
+1. **Add the model without provider**:
+
+```json
+{
+  "models": [
+    {
+      "hrid": "claude-3-opus",
+      "model_name": "anthropic:claude-3-opus-20240229",
+      "human_readable_name": "Claude 3 Opus",
+      "provider_name": null,
+      "profile": null,
+      "settings": {
+        "temperature": 0.7,
+        "max_tokens": 4096
+      },
+      "is_active": true,
+      "icon": null,
+      "system_prompt": "You are Claude, a helpful AI assistant.",
+      "tools": []
+    }
+  ],
+  "providers": []
+}
+```
+
+2**Set the environment variable**:
+
+Pydantic AI expects the API key in an environment variable named `ANTHROPIC_API_KEY` is this example, so set it accordingly:
+
+```ini
+ANTHROPIC_API_KEY=your-api-key-here
+```
+
+### Example 3: Adding a Mistral Model
+
+For Mistral AI models using the Etalab platform:
+
+```json
+{
+  "models": [
+    {
+      "hrid": "mistral-medium",
+      "model_name": "mistral-medium-2508",
+      "human_readable_name": "Mistral Medium (Etalab)",
+      "provider_name": "mistral-etalab",
+      "profile": null,
+      "settings": {
+        "temperature": 0.5,
+        "max_tokens": 8192
+      },
+      "is_active": true,
+      "icon": null,
+      "system_prompt": "settings.AI_AGENT_INSTRUCTIONS",
+      "tools": ["web_search_brave_with_document_backend"]
+    }
+  ],
+  "providers": [
+    {
+      "hrid": "mistral-etalab",
+      "base_url": "https://api.mistral.etalab.gouv.fr/",
+      "api_key": "environ.MISTRAL_ETALAB_API_KEY",
+      "kind": "mistral"
+    }
+  ]
+}
+```
+
+### Example 4: Using Inline Provider Definition
+
+Instead of referencing a provider by name, you can define it inline if you use a unique configuration:
+
+```json
+{
+  "models": [
+    {
+      "hrid": "custom-model",
+      "model_name": "custom-model-v1",
+      "human_readable_name": "Custom Model",
+      "provider": {
+        "hrid": "custom-provider-inline",
+        "base_url": "https://custom-api.example.com/v1",
+        "api_key": "environ.CUSTOM_API_KEY",
+        "kind": "openai"
+      },
+      "settings": {},
+      "is_active": true,
+      "icon": null,
+      "system_prompt": "You are a custom assistant.",
+      "tools": []
+    }
+  ]
+}
+```
+
+## Advanced Configuration
+
+### Model Settings
+
+The `settings` object supports various inference parameters:
+
+```json
+{
+  "settings": {
+    "max_tokens": 4096,
+    "temperature": 0.7,
+    "top_p": 0.9,
+    "timeout": 60.0,
+    "parallel_tool_calls": true,
+    "seed": 42,
+    "presence_penalty": 0.0,
+    "frequency_penalty": 0.0,
+    "logit_bias": {},
+    "stop_sequences": [],
+    "extra_headers": {},
+    "extra_body": {}
+  }
+}
+```
+
+### Model Profile
+
+The `profile` object defines model capabilities:
+
+```json
+{
+  "profile": {
+    "supports_tools": true,
+    "supports_json_schema_output": true,
+    "supports_json_object_output": true,
+    "default_structured_output_mode": "json_schema",
+    "thinking_tags": ["<thinking>", "</thinking>"],
+    "ignore_streamed_leading_whitespace": true
+  }
+}
+```
+
+### Available Tools
+
+Tools can be specified in the `tools` array. Common tools include:
+- `web_search_brave_with_document_backend`: Web search using Brave API with document processing
+
+You can also reference the tools list from Django settings:
+```json
+{
+  "tools": "settings.AI_AGENT_TOOLS"
+}
+```
+
+### Custom Icons
+
+Icons can be provided as base64-encoded PNG images. For long strings, you can split them into an array:
+
+```json
+{
+  "icon": [
+    "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABwAAAAcCAMAAABF0y+m",
+    "AAAAn1BMVEUALosAKoovTZjw8vb////+9/jlPUniAAziABUAGIWbpsTwq7HhAAAA"
+  ]
+}
+```
+
+## Validation
+
+The configuration is validated when loaded. Common validation errors include:
+
+- **Provider not found**: A model references a `provider_name` that doesn't exist in the `providers` array
+- **Missing provider**: Neither `provider_name` nor `provider` is specified, and `model_name` is not in `<provider>:<model>` format
+- **Environment variable not set**: A value using `environ.` prefix references an undefined environment variable
+- **Django setting not set**: A value using `settings.` prefix references an undefined Django setting
+- **Invalid provider kind**: The `kind` field must be either `openai` or `mistral`
+
+## Testing Your Configuration
+
+After modifying the configuration file, you can test it by:
+
+1. **Checking for syntax errors**:
+   ```bash
+   python -m json.tool src/backend/conversations/configuration/llm/default.json
+   ```
+
+2. **Starting the application** and checking the logs for validation errors
+
+3. **Using the Django shell** to load the configuration:
+   ```bash
+   ./bin/manage shell
+   ```
+   ```python
+   from django.conf import settings
+   models = settings.LLM_CONFIGURATIONS
+   models.keys()  # Should show all model HRIDs
+   ```
+
+## Best Practices
+
+1. **Use environment variables** for sensitive data like API keys (with `environ.` prefix)
+2. **Use Django settings** for configurable values that may change between environments (with `settings.` prefix)
+3. **Keep provider definitions separate** from models to avoid duplication when using multiple models from the same provider
+4. **Set `is_active: false`** for models you want to keep in the configuration but temporarily disable
+5. **Use descriptive `hrid` values** that clearly identify the model and provider
+6. **Document custom configurations** in your deployment documentation
+7. **Test configuration changes** in a development environment before deploying to production
+
+## See Also
+
+- [Environment Variables Documentation](env.md) - For configuring environment variables
+- [Installation Guide](installation.md) - For deployment instructions
+

docs/system-requirements.md

@@ -1,9 +1,9 @@
-# La Suite Docs – System & Requirements (2025-06)
+# La Suite Conversations – System & Requirements (2025-06)
 
 ## 1. Quick-Reference Matrix (single VM / laptop)
 
 | Scenario                  | RAM   | vCPU | SSD     | Notes                     |
-| ------------------------- | ----- | ---- | ------- | ------------------------- |
+|---------------------------|-------|------|---------|---------------------------|
 | **Solo dev**              | 8 GB  | 4    | 15 GB   | Hot-reload + one IDE      |
 | **Team QA**               | 16 GB | 6    | 30 GB   | Runs integration tests    |
 | **Prod ≤ 100 live users** | 32 GB | 8 +  | 50 GB + | Scale linearly above this |
@@ -14,22 +14,20 @@ Memory is the first bottleneck; CPU matters only when Celery or the Next.js buil
 
 ## 2. Development Environment Memory Requirements
 
-| Service                  | Typical use                   | Rationale / source                                                                      |
-| ------------------------ | ----------------------------- | --------------------------------------------------------------------------------------- |
-| PostgreSQL               | **1 – 2 GB**                 | `shared_buffers` starting point ≈ 25% RAM ([postgresql.org][1])                       |
-| Keycloak                 | **≈ 1.3 GB**                 | 70% of limit for heap + ~300 MB non-heap ([keycloak.org][2])                         |
-| Redis                    | **≤ 256 MB**                 | Empty instance ≈ 3 MB; budget 256 MB to allow small datasets ([stackoverflow.com][3]) |
-| MinIO                    | **2 GB (dev) / 32 GB (prod)**| Pre-allocates 1–2 GiB; docs recommend 32 GB per host for ≤ 100 Ti storage ([min.io][4]) |
-| Django API (+ Celery)    | **0.8 – 1.5 GB**              | Empirical in-house metrics                                                              |
-| Next.js frontend         | **0.5 – 1 GB**                | Dev build chain                                                                         |
-| Y-Provider (y-websocket) | **< 200 MB**                  | Large 40 MB YDoc called “big” in community thread ([discuss.yjs.dev][5])                |
-| Nginx                    | **< 100 MB**                  | Static reverse-proxy footprint                                                          |
+| Service          | Typical use                   | Rationale / source                                                                      |
+|------------------|-------------------------------|-----------------------------------------------------------------------------------------|
+| PostgreSQL       | **1 – 2 GB**                  | `shared_buffers` starting point ≈ 25% RAM ([postgresql.org][1])                         |
+| Keycloak         | **≈ 1.3 GB**                  | 70% of limit for heap + ~300 MB non-heap ([keycloak.org][2])                            |
+| Redis            | **≤ 256 MB**                  | Empty instance ≈ 3 MB; budget 256 MB to allow small datasets ([stackoverflow.com][3])   |
+| MinIO            | **2 GB (dev) / 32 GB (prod)** | Pre-allocates 1–2 GiB; docs recommend 32 GB per host for ≤ 100 Ti storage ([min.io][4]) |
+| Django API       | **0.8 – 1.5 GB**              | Empirical in-house metrics                                                              |
+| Next.js frontend | **0.5 – 1 GB**                | Dev build chain                                                                         |
+| Nginx            | **< 100 MB**                  | Static reverse-proxy footprint                                                          |
 
 [1]: https://www.postgresql.org/docs/9.1/runtime-config-resource.html "PostgreSQL: Documentation: 9.1: Resource Consumption"
 [2]: https://www.keycloak.org/high-availability/concepts-memory-and-cpu-sizing "Concepts for sizing CPU and memory resources - Keycloak"
 [3]: https://stackoverflow.com/questions/45233052/memory-footprint-for-redis-empty-instance "Memory footprint for Redis empty instance - Stack Overflow"
 [4]: https://min.io/docs/minio/kubernetes/upstream/operations/checklists/hardware.html "Hardware Checklist — MinIO Object Storage for Kubernetes"
-[5]: https://discuss.yjs.dev/t/understanding-memory-requirements-for-production-usage/198 "Understanding memory requirements for production usage - Yjs Community"
 
 > **Rule of thumb:** add 2 GB for OS/overhead, then sum only the rows you actually run.
 
@@ -37,16 +35,15 @@ Memory is the first bottleneck; CPU matters only when Celery or the Next.js buil
 
 Production deployments differ significantly from development environments. The table below shows typical memory usage for production services:
 
-| Service                  | Typical use                   | Rationale / notes                                                                       |
-| ------------------------ | ----------------------------- | --------------------------------------------------------------------------------------- |
-| PostgreSQL               | **2 – 8 GB**                 | Higher `shared_buffers` and connection pooling for concurrent users                    |
-| OIDC Provider (optional) | **Variable**                  | Any OIDC-compatible provider (Keycloak, Auth0, Azure AD, etc.) - external or self-hosted |
-| Redis                    | **256 MB – 2 GB**             | Session storage and caching; scales with active user sessions                          |
-| Object Storage (optional)| **External or self-hosted**   | Can use AWS S3, Azure Blob, Google Cloud Storage, or self-hosted MinIO               |
-| Django API (+ Celery)    | **1 – 3 GB**                 | Production workloads with background tasks and higher concurrency                      |
-| Static Files (Nginx)     | **< 200 MB**                 | Serves Next.js build output and static assets; no development overhead                |
-| Y-Provider (y-websocket) | **200 MB – 1 GB**             | Scales with concurrent document editing sessions                                        |
-| Nginx (Load Balancer)    | **< 200 MB**                  | Reverse proxy, SSL termination, static file serving                                    |
+| Service                   | Typical use                 | Rationale / notes                                                                        |
+|---------------------------|-----------------------------|------------------------------------------------------------------------------------------|
+| PostgreSQL                | **2 – 8 GB**                | Higher `shared_buffers` and connection pooling for concurrent users                      |
+| OIDC Provider (optional)  | **Variable**                | Any OIDC-compatible provider (Keycloak, Auth0, Azure AD, etc.) - external or self-hosted |
+| Redis                     | **256 MB – 2 GB**           | Session storage and caching; scales with active user sessions                            |
+| Object Storage (optional) | **External or self-hosted** | Can use AWS S3, Azure Blob, Google Cloud Storage, or self-hosted MinIO                   |
+| Django API (+ Celery)     | **1 – 3 GB**                | Production workloads with background tasks and higher concurrency                        |
+| Static Files (Nginx)      | **< 200 MB**                | Serves Next.js build output and static assets; no development overhead                   |
+| Nginx (Load Balancer)     | **< 200 MB**                | Reverse proxy, SSL termination, static file serving                                      |
 
 ### Production Architecture Notes
 
@@ -54,14 +51,14 @@ Production deployments differ significantly from development environments. The t
 - **Authentication**: Any OIDC-compatible provider can be used instead of self-hosted Keycloak
 - **Object Storage**: External services (S3, Azure Blob) or self-hosted solutions (MinIO) are both viable
 - **Database**: Consider PostgreSQL clustering or managed database services for high availability
-- **Scaling**: Horizontal scaling is recommended for Django API and Y-Provider services
+- **Scaling**: Horizontal scaling is recommended for Django API service
 
 ### Minimal Production Setup (Core Services Only)
 
 | Service                          | Memory     | Notes                                  |
 |----------------------------------|------------|----------------------------------------|
 | PostgreSQL                       | **2 GB**   | Core database                          |
-| Django API (+ Celery)            | **1.5 GB** | Backend services                       |
+| Django API                       | **1.5 GB** | Backend services                       |
 | Nginx                            | **100 MB** | Static files + reverse proxy           |
 | Redis                            | **256 MB** | Session storage                        |
 | **Total (without auth/storage)** | **≈ 4 GB** | External OIDC + object storage assumed |
@@ -69,7 +66,7 @@ Production deployments differ significantly from development environments. The t
 ## 4. Recommended Software Versions
 
 | Tool                    | Minimum |
-| ----------------------- | ------- |
+|-------------------------|---------|
 | Docker Engine / Desktop | 24.0    |
 | Docker Compose          | v2      |
 | Git                     | 2.40    |
@@ -84,17 +81,16 @@ Production deployments differ significantly from development environments. The t
 
 ## 5. Ports (dev defaults)
 
-| Port      | Service               |
-| --------- |-----------------------|
-| 3000      | Next.js               |
-| 8071      | Django                |
-| 4444      | Y-Provider            |
-| 8080      | Keycloak              |
-| 8083      | Nginx proxy           |
-| 9000/9001 | MinIO                 |
-| 15432     | PostgreSQL (main)     |
-| 5433      | PostgreSQL (Keycloak) |
-| 1081      | Maildev               |
+| Port      | Service                    |
+|-----------|----------------------------|
+| 3000      | Next.js                    |
+| 8071      | Django                     |
+| 8080      | Keycloak                   |
+| 8083      | Nginx proxy                |
+| 9000/9001 | MinIO                      |
+| 15432     | PostgreSQL (main)          |
+| 5433      | PostgreSQL (Keycloak)      |
+| 1081      | Maildev (currently unused) |
 
 ## 6. Sizing Guidelines
 
@@ -106,4 +102,4 @@ Production deployments differ significantly from development environments. The t
 
 **Disk** – SSD; add 10 GB extra for the Docker layer cache.
 
-**MinIO** – for demos, mount a local folder instead of running MinIO to save 2 GB+ of RAM.
+**MinIO** – for demos, mount a local folder instead of running MinIO to save 2 GB+ of RAM.

docs/theming.md

@@ -4,7 +4,7 @@
 
 To use this feature, simply set the `FRONTEND_CSS_URL` environment variable to the URL of your custom CSS file. For example:
 
-```javascript
+```ini
 FRONTEND_CSS_URL=http://anything/custom-style.css
 ```
 
@@ -38,7 +38,7 @@ The footer is configurable from the theme customization file.
 
 ### Settings 🔧
 
-```shellscript
+```ini
 THEME_CUSTOMIZATION_FILE_PATH=<path>
 ```
 
@@ -55,10 +55,10 @@ The translations can be partially overridden from the theme customization file.
 
 ### Settings 🔧
 
-```shellscript
+```ini
 THEME_CUSTOMIZATION_FILE_PATH=<path>
 ```
 
 ### Example of JSON
 
-The json must follow some rules: https://github.com/suitenumerique/conversations/blob/main/src/helm/env.d/dev/configuration/theme/demo.json
+The json must follow some rules: https://github.com/suitenumerique/conversations/blob/main/src/helm/env.d/dev/configuration/theme/demo.json

docs/tools.md

@@ -0,0 +1,238 @@
+# Tools for the Conversation Agent
+
+The conversation agent can be extended with various tools that provide additional capabilities such as web search, 
+weather information, and more. We currently only have web search tools, but more tools can be added as needed.
+This document explains how to configure and use these tools.
+
+## Overview
+
+Tools are functions that the LLM can call during a conversation to access external data or perform specific actions. 
+The agent decides when to use these tools based on the user's query and the conversation context.
+
+## Configuring Tools for a Model
+
+Tools are configured at the model level in the LLM configuration file. 
+Each model can have its own set of available tools.
+
+### Configuration File Location
+
+Read the [LLM Configuration](llm-configuration.md) document to find out where the configuration file is located
+and how to use it.
+
+### Example Configuration
+
+```json
+{
+  "models": [
+    {
+      "hrid": "default-model",
+      "model_name": "gpt-4",
+      "human_readable_name": "GPT-4 with Tools",
+      "provider_name": "default-provider",
+      "is_active": true,
+      "system_prompt": "You are a helpful assistant.",
+      "tools": [
+        "web_search_brave",
+        "get_current_weather"
+      ]
+    }
+  ],
+  "providers": [
+    {
+      "hrid": "default-provider",
+      "base_url": "https://api.openai.com/v1",
+      "api_key": "settings.AI_API_KEY",
+      "kind": "openai"
+    }
+  ]
+}
+```
+
+The `tools` field accepts either:
+- A list of tool names: `["tool_name_1", "tool_name_2"]`
+- A reference to a settings variable: `"settings.AI_AGENT_TOOLS"`
+
+## Available Tools
+
+To make a tool available to be in a model's configuration, it must be registered in the tool registry located at
+`src/backend/chat/tools/__init__.py`.
+
+This is not dynamic - any changes to the tool registry require a code deployment... 
+We want to add dynamic loading in the future.
+
+| Tool Name                                | Description                                                   | Documentation                                                               |
+|------------------------------------------|---------------------------------------------------------------|-----------------------------------------------------------------------------|
+| `get_current_weather`                    | Fake weather tool for testing purposes                        | [Details](tools/get_current_weather.md)                                     |
+| `web_search_tavily`                      | Web search using Tavily API                                   | [Details](tools/web_search_tavily.md)                                       |
+| `web_search_brave`                       | Web search using Brave Search API with optional summarization | [Details](tools/web_search_brave.md)                                        |
+| `web_search_brave_with_document_backend` | Web search using Brave with RAG-based document processing     | [Details](tools/web_search_brave.md#web_search_brave_with_document_backend) |
+| `web_search_albert_rag`                  | ⚠️ **Deprecated** - Web search using Albert API with RAG      | [Details](tools/web_search_brave.md#deprecated-web_search_albert_rag)       |
+
+## Adding a New Tool
+
+To add a new tool to the system, follow these steps:
+
+### 1. Create the Tool Function
+
+Create a new Python file in `src/backend/chat/tools/` with your tool function. The function should:
+
+- Have clear type annotations
+- Include a comprehensive docstring (the LLM uses this to understand when to use the tool)
+- Accept `RunContext` as the first parameter if it needs access to conversation context
+- Return appropriate data types
+
+Example:
+```python
+"""My custom tool for the chat agent."""
+
+from pydantic_ai import RunContext
+
+def my_custom_tool(ctx: RunContext, param1: str, param2: int) -> dict:
+    """
+    Brief description of what the tool does.
+    
+    The LLM uses this description to decide when to call this tool.
+    
+    Args:
+        ctx (RunContext): The run context containing the conversation.
+        param1 (str): Description of parameter 1.
+        param2 (int): Description of parameter 2.
+    
+    Returns:
+        dict: Description of the return value.
+    """
+    # Your implementation here
+    return {"result": "example"}
+```
+
+### 2. Register the Tool
+
+Add your tool to the registry in `src/backend/chat/tools/__init__.py`:
+
+```python
+from .my_custom_tool import my_custom_tool
+
+def get_pydantic_tools_by_name(name: str) -> Tool:
+    """Get a tool by its name."""
+    tool_dict = {
+        "get_current_weather": Tool(get_current_weather, takes_ctx=False),
+        "web_search_brave": Tool(
+            web_search_brave, takes_ctx=False, prepare=only_if_web_search_enabled
+        ),
+        # Add your tool here
+        "my_custom_tool": Tool(
+            my_custom_tool, 
+            takes_ctx=True,  # Set to True if your tool needs RunContext
+            # prepare=only_if_web_search_enabled  # Optional: add conditions
+        ),
+    }
+    return tool_dict[name]
+```
+
+### 3. Update Imports
+
+Don't forget to import your tool function at the top of `__init__.py`:
+
+```python
+from .my_custom_tool import my_custom_tool
+```
+
+### 4. Add to Model Configuration
+
+Add your tool name to the `tools` list in your LLM configuration file or 
+to the `AI_AGENT_TOOLS` environment variable for local/test purpose.
+
+## Tool Preparation: Conditional Tool Availability
+
+Some tools should only be available under certain conditions. The `prepare` parameter in the `Tool` constructor 
+allows you to specify a function that determines whether a tool should be included.
+
+### The `only_if_web_search_enabled` Prepare Function
+
+This is a built-in prepare function that checks if web search feature is enabled in the conversation context:
+
+```python
+async def only_if_web_search_enabled(ctx, tool_def: ToolDefinition) -> ToolDefinition | None:
+    """Prepare function to include a tool only if web search is enabled in the context."""
+    return tool_def if ctx.deps.web_search_enabled else None
+```
+
+### Usage
+
+All web search tools use this prepare function:
+
+```python
+"web_search_brave": Tool(
+    web_search_brave, 
+    takes_ctx=False, 
+    prepare=only_if_web_search_enabled
+),
+```
+
+This ensures that web search tools are only available when the user or conversation settings have enabled web search functionality.
+
+### Creating Custom Prepare Functions
+
+You can create your own prepare functions for custom conditions:
+
+```python
+async def only_if_feature_enabled(ctx, tool_def: ToolDefinition) -> ToolDefinition | None:
+    """Include tool only if a specific feature is enabled."""
+    return tool_def if ctx.deps.feature_enabled else None
+```
+
+## Web Search Enable/Disable
+
+Web search tools can be toggled on or off based on conversation settings. When web search is disabled:
+- Web search tools are not included in the agent's available tools
+- The LLM cannot make web search calls even if it tries
+- This is enforced by the `only_if_web_search_enabled` prepare function
+
+The `web_search_enabled` flag is typically set:
+- Per conversation in the conversation settings
+- Per user preference
+- Through admin configuration
+
+## Best Practices
+
+1. **Keep tools focused** - Each tool should do one thing well
+2. **Clear documentation** - The LLM relies on docstrings to understand when to use tools
+3. **Error handling** - Tools should handle errors gracefully and return meaningful messages
+4. **Performance** - Be mindful of API rate limits and timeout values
+5. **Security** - Never log sensitive data (API keys, user data, etc.)
+6. **Caching** - Use Django's cache framework for expensive operations when appropriate
+
+## Troubleshooting
+
+### Tool Not Being Called
+
+If the LLM isn't calling your tool:
+- Check that the tool is registered in `get_pydantic_tools_by_name`
+- Verify the tool is in the model's `tools` configuration
+- Review the tool's docstring - make it clearer when the tool should be used
+- Check if any `prepare` function is preventing the tool from being included
+
+### Tool Errors
+
+If a tool is throwing errors:
+- Check the logs for detailed error messages
+- Verify all required environment variables are set
+- Ensure the tool's dependencies are installed
+- Test the tool function independently
+
+We recommend wrapping external API calls in try/except blocks to handle potential issues gracefully and use
+the Pydantic AI `ModelRetry` exception to let the LLM manage the errors.
+
+### Tool Response Issues
+
+If the LLM isn't using the tool response correctly:
+- Ensure the return type is clear and well-structured
+- Consider returning a `ToolReturn` object with metadata
+- Check if the response format matches what the LLM expects
+
+## See Also
+
+- [Web Search Configuration](llm-configuration.md)
+- [Architecture](architecture.md)
+- [Environment Variables](env.md)
+

docs/tools/get_current_weather.md

@@ -0,0 +1,113 @@
+# get_current_weather Tool
+
+## Overview
+
+The `get_current_weather` tool is a **fake weather tool** designed for testing and demonstration purposes. It does not connect to any real weather API and always returns hardcoded weather data.
+
+## Purpose
+
+This tool is useful for:
+- **Testing** the tool calling functionality of LLMs
+- **Demonstrating** how tools work without requiring API keys
+- **Development** and debugging of the agent system
+- **Example implementation** for creating new tools
+
+⚠️ **Warning**: This tool should **not** be used in production environments. It always returns fake data regardless of the location or conditions.
+
+## Configuration
+
+### Add to Model
+
+To enable this tool for a model, add it to the `tools` list in your LLM configuration:
+
+```json
+{
+  "models": [
+    {
+      "hrid": "my-model",
+      "tools": [
+        "get_current_weather"
+      ]
+    }
+  ]
+}
+```
+
+Or via environment variable when using local environment settings:
+```ini
+AI_AGENT_TOOLS=get_current_weather
+```
+
+### No Additional Settings Required
+
+This tool does not require any API keys, environment variables, or additional configuration.
+
+## Function Signature
+
+```python
+def get_current_weather(location: str, unit: str) -> dict:
+    """
+    Get the current weather in a given location.
+
+    Args:
+        location (str): The city and state, e.g. San Francisco, CA.
+        unit (str): The unit of temperature, either 'celsius' or 'fahrenheit'.
+
+    Returns:
+        dict: A dictionary containing the location, temperature, and unit.
+    """
+```
+
+## Parameters
+
+| Parameter  | Type | Required | Description                                                     |
+|------------|------|----------|-----------------------------------------------------------------|
+| `location` | str  | Yes      | The city and state (e.g., "San Francisco, CA", "Paris, France") |
+| `unit`     | str  | Yes      | Temperature unit: either "celsius" or "fahrenheit"              |
+
+## Return Value
+
+Returns a dictionary with the following structure:
+
+```python
+{
+    "location": str,      # The location that was queried
+    "temperature": int,   # Always 22°C or 72°F
+    "unit": str          # The unit that was requested
+}
+```
+
+## How the LLM Uses It
+
+When a user asks about weather, the LLM will:
+
+1. **Recognize** the weather-related query
+2. **Extract** the location from the user's message
+3. **Determine** the appropriate unit (often from context or user preference)
+4. **Call** the `get_current_weather` tool
+5. **Receive** the fake weather data
+6. **Format** a response to the user
+
+### Example Conversation
+
+**User**: "What's the weather like in London?"
+
+**LLM** (internal): *Calls `get_current_weather("London, UK", "celsius")`*
+
+**Tool Response**: 
+```json
+{
+  "location": "London, UK",
+  "temperature": 22,
+  "unit": "celsius"
+}
+```
+
+**LLM** (to user): "The current weather in London, UK is 22°C."
+
+## See Also
+
+- [Tools Overview](../tools.md)
+- [Adding a New Tool](../tools.md#adding-a-new-tool)
+- [Testing Tools](../tools.md#testing-your-tools)
+

docs/tools/web_search_brave.md

@@ -0,0 +1,671 @@
+# Brave Web Search Tools
+
+## Overview
+
+The Brave web search tools enable the conversation agent to search the web using the [Brave Search API](https://brave.com/search/api/). 
+Brave Search is a privacy-focused search engine that provides comprehensive web search results.
+
+This documentation covers three related tools:
+1. **`web_search_brave`** - Standard web search with optional summarization
+2. **`web_search_brave_with_document_backend`** - Web search with RAG-based document processing
+3. **`web_search_albert_rag`** - ⚠️ **Deprecated** - Use `web_search_brave_with_document_backend` instead
+
+## Table of Contents
+
+- [Common Configuration](#common-configuration)
+- [web_search_brave](#web_search_brave)
+- [web_search_brave_with_document_backend](#web_search_brave_with_document_backend)
+- [Deprecated: web_search_albert_rag](#deprecated-web_search_albert_rag)
+- [Comparison](#comparison)
+- [Best Practices](#best-practices)
+- [Troubleshooting](#troubleshooting)
+
+---
+
+## Common Configuration
+
+### Prerequisites
+
+1. **Brave Search API Key**: Sign up at [Brave Search API](https://brave.com/search/api/) to get an API key
+2. **Environment Variables**: Configure the required settings
+
+### Common Environment Variables
+
+All Brave tools share these common settings:
+
+| Variable            | Required | Default | Description                                        |
+|---------------------|----------|---------|----------------------------------------------------|
+| `BRAVE_API_KEY`     | **Yes**  | None    | Your Brave Search API key                          |
+| `BRAVE_API_TIMEOUT` | No       | 5       | API request timeout in seconds                     |
+| `BRAVE_MAX_RESULTS` | No       | 8       | Maximum number of search results                   |
+| `BRAVE_CACHE_TTL`   | No       | 1800    | Cache time-to-live in seconds (30 minutes)         |
+
+### Search Parameters
+
+Check on the Brave API documentation for more details on these parameters:
+
+| Variable                      | Required | Default    | Description                                       |
+|-------------------------------|----------|------------|---------------------------------------------------|
+| `BRAVE_SEARCH_COUNTRY`        | No       | None       | Country code for search (e.g., "US", "FR")        |
+| `BRAVE_SEARCH_LANG`           | No       | None       | Language code (e.g., "en", "fr")                  |
+| `BRAVE_SEARCH_SAFE_SEARCH`    | No       | "moderate" | Safe search level: "off", "moderate", or "strict" |
+| `BRAVE_SEARCH_SPELLCHECK`     | No       | True       | Enable spell checking                             |
+| `BRAVE_SEARCH_EXTRA_SNIPPETS` | No       | True       | Fetch extra snippets from pages                   |
+
+
+Note: even if `BRAVE_SEARCH_EXTRA_SNIPPETS` is enabled, the API may not include them if you don't have a plan for this.
+This is why, in `web_search_brave`, we also fetch the page content ourselves when needed.
+
+### Configuration Example
+
+```bash
+# .env file
+BRAVE_API_KEY=BSA-your-api-key-here
+BRAVE_MAX_RESULTS=8
+BRAVE_MAX_WORKERS=4
+BRAVE_SEARCH_COUNTRY=US
+BRAVE_SEARCH_LANG=en
+BRAVE_SEARCH_SAFE_SEARCH=moderate
+```
+
+### Django Settings
+
+All Brave settings are defined in `src/backend/conversations/brave_settings.py`:
+
+```python
+class BraveSettings:
+    """Brave settings for web_search_brave tool."""
+
+    BRAVE_API_KEY = values.Value(
+        default=None,
+        environ_name="BRAVE_API_KEY",
+        environ_prefix=None,
+    )
+    # ... more settings
+```
+
+---
+
+## web_search_brave
+
+### Overview
+
+Standard Brave web search tool with optional LLM-based summarization of page content.
+
+### Purpose
+
+- Search the web for up-to-date information
+- Extract content from web pages
+- Optionally summarize content using an LLM
+- Provide structured results with snippets
+
+### Additional Configuration
+
+| Variable                      | Required | Default | Description                                     |
+|-------------------------------|----------|---------|-------------------------------------------------|
+| `BRAVE_SUMMARIZATION_ENABLED` | No       | False   | Enable LLM-based summarization of fetched pages |
+
+### Function Signature
+
+```python
+def web_search_brave(query: str) -> ToolReturn:
+    """
+    Search the web for up-to-date information
+
+    Args:
+        query (str): The query to search for.
+        
+    Returns:
+        ToolReturn: Formatted search results with metadata
+    """
+```
+
+### Return Value
+
+Returns a `ToolReturn` object with:
+
+```python
+ToolReturn(
+    return_value={
+        "0": {
+            "url": "https://example.com/page1",
+            "title": "Example Page Title",
+            "snippets": ["Extracted or summarized content..."]
+        },
+        "1": {
+            "url": "https://example.com/page2",
+            "title": "Another Page",
+            "snippets": ["More content..."]
+        }
+    },
+    metadata={
+        "sources": {
+            "https://example.com/page1",
+            "https://example.com/page2"
+        }
+    }
+)
+```
+
+### How It Works
+
+1. **Query API**: Sends search query to Brave Search API
+2. **Receive Results**: Gets list of matching web pages
+3. **Fetch Content**: For results without extra_snippets:
+   - Fetches the HTML content using `trafilatura`
+   - Extracts the main text content
+   - Caches the extracted content
+4. **Summarize (Optional)**: If `BRAVE_SUMMARIZATION_ENABLED=True`:
+   - Sends extracted content to summarization agent
+   - Receives concise summary focused on the query
+5. **Format Results**: Returns structured data with URLs, titles, and snippets
+
+### Workflow Diagram
+
+```
+User Query
+    ↓
+Brave Search API
+    ↓
+Search Results (URLs, titles, descriptions)
+    ↓
+[For each result without snippets]
+    ↓
+Fetch HTML (trafilatura) → Extract Text → Cache
+    ↓
+[If BRAVE_SUMMARIZATION_ENABLED]
+    ↓
+Summarization Agent (LLM)
+    ↓
+Summary Text
+    ↓
+Format & Return
+```
+
+### Caching
+
+Extracted content is cached to avoid repeated fetches:
+
+```python
+cache_key = f"web_search_brave:extract:{url}"
+cache.set(cache_key, document, settings.BRAVE_CACHE_TTL)
+```
+
+**Cache Duration**: Controlled by `BRAVE_CACHE_TTL` (default: 30 minutes)
+
+### Summarization
+
+When enabled, the tool uses the `SummarizationAgent` to condense page content:
+
+```python
+prompt = f"""
+Based on the following request, summarize the following text in a concise manner, 
+focusing on the key points regarding the user request. 
+The result should be up to 30 lines long.
+
+<user request>
+{query}
+</user request>
+
+<text to summarize>
+{text}
+</text to summarize>
+"""
+```
+
+**Note**: Summarization is costly (additional LLM calls). 
+Use only when necessary, we prefer the document vector search from `web_search_brave_with_document_backend`.
+
+### Add to Model
+
+```json
+{
+  "models": [
+    {
+      "hrid": "my-model",
+      "tools": [
+        "web_search_brave"
+      ]
+    }
+  ]
+}
+```
+
+### Example Usage
+
+**User**: "What are the new features in Django 5.0?"
+
+**Tool Call**: `web_search_brave("Django 5.0 new features")`
+
+**Tool Response**:
+```python
+{
+    "0": {
+        "url": "https://docs.djangoproject.com/en/5.0/releases/5.0/",
+        "title": "Django 5.0 release notes",
+        "snippets": ["Django 5.0 introduces several new features including..."]
+    },
+    # ... more results
+}
+```
+
+### Registration
+
+```python
+"web_search_brave": Tool(
+    web_search_brave, 
+    takes_ctx=False, 
+    prepare=only_if_web_search_enabled
+)
+```
+
+---
+
+## web_search_brave_with_document_backend
+
+### Overview
+
+Advanced Brave web search tool that uses RAG (Retrieval-Augmented Generation) 
+with a document backend for intelligent content processing and retrieval.
+
+### Purpose
+
+- Search the web and process results through a RAG system
+- Store fetched documents in a temporary vector database
+- Perform semantic search across fetched content
+- Return the most relevant chunks based on the query
+
+### Additional Configuration
+
+| Variable                            | Required | Default          | Description                                  |
+|-------------------------------------|----------|------------------|----------------------------------------------|
+| `BRAVE_RAG_WEB_SEARCH_CHUNK_NUMBER` | No       | 10               | Number of chunks to retrieve from RAG search |
+| `RAG_DOCUMENT_SEARCH_BACKEND`       | No       | AlbertRagBackend | Document backend for RAG processing          |
+
+### Function Signature
+
+```python
+def web_search_brave_with_document_backend(ctx: RunContext, query: str) -> ToolReturn:
+    """
+    Search the web for up-to-date information
+
+    Args:
+        ctx (RunContext): The run context containing the conversation.
+        query (str): The query to search for.
+        
+    Returns:
+        ToolReturn: Formatted search results with RAG-enhanced snippets
+    """
+```
+
+### How It Works
+
+1. **Query API**: Sends search query to Brave Search API
+2. **Receive Results**: Gets list of matching web pages
+3. **Create Temporary Collection**: Creates a temporary vector database collection
+4. **Fetch & Store**: For each result:
+   - Fetches the HTML content
+   - Extracts the main text
+   - Stores in the temporary document backend
+5. **RAG Search**: Performs semantic search across stored documents
+6. **Map Results**: Maps RAG chunks back to original search results
+7. **Format & Return**: Returns structured data with enhanced snippets
+8. **Cleanup**: Temporary collection is automatically deleted
+
+### Workflow Diagram
+
+```
+User Query
+    ↓
+Brave Search API
+    ↓
+Search Results (URLs)
+    ↓
+Create Temporary Vector Collection
+    ↓
+[For each URL]
+    ↓
+Fetch HTML → Extract Text → Store in Vector DB
+    ↓
+RAG Semantic Search
+    ↓
+Retrieve Most Relevant Chunks
+    ↓
+Map Chunks to Original URLs
+    ↓
+Format & Return
+    ↓
+Delete Temporary Collection
+```
+
+### Temporary Collection
+
+The tool creates a temporary collection with a unique ID:
+
+```python
+with document_store_backend.temporary_collection(f"tmp-{uuid.uuid4()}") as document_store:
+    # Fetch and store documents
+    # Perform search
+    # Collection is automatically deleted on exit
+```
+
+### RAG Search
+
+The RAG backend performs semantic search to find the most relevant content:
+
+```python
+rag_results = document_store.search(
+    query,
+    results_count=settings.BRAVE_RAG_WEB_SEARCH_CHUNK_NUMBER,
+    **kwargs, # Additional search parameters like session with access_token
+)
+```
+
+Returns chunks ranked by relevance to the query, not just keyword matching.
+
+### Token Usage Tracking
+
+The tool tracks LLM tokens used during RAG processing:
+
+```python
+ctx.usage += RunUsage(
+    input_tokens=rag_results.usage.prompt_tokens,
+    output_tokens=rag_results.usage.completion_tokens,
+)
+```
+
+### Document Backend
+
+The default backend is `AlbertRagBackend`, but you can configure a different one:
+
+```bash
+RAG_DOCUMENT_SEARCH_BACKEND=chat.agent_rag.document_rag_backends.custom_backend.CustomBackend
+```
+
+### Add to Model
+
+```json
+{
+  "models": [
+    {
+      "hrid": "my-model",
+      "tools": [
+        "web_search_brave_with_document_backend"
+      ]
+    }
+  ]
+}
+```
+
+### Example Usage
+
+**User**: "Explain the concept of async views in Django"
+
+**Tool Call**: `web_search_brave_with_document_backend(ctx, "Django async views explained")`
+
+**Tool Response**:
+```python
+{
+    "0": {
+        "url": "https://docs.djangoproject.com/en/stable/topics/async/",
+        "title": "Asynchronous support",
+        "snippets": [
+            "Django has support for writing asynchronous views...",
+            "Async views are declared using Python's async def syntax..."
+        ]
+    },
+    # ... more results with relevant chunks
+}
+```
+
+### Registration
+
+```python
+"web_search_brave_with_document_backend": Tool(
+    web_search_brave_with_document_backend,
+    takes_ctx=True,
+    prepare=only_if_web_search_enabled,
+)
+```
+
+### Advantages Over Standard web_search_brave
+
+| Feature           | web_search_brave               | web_search_brave_with_document_backend |
+|-------------------|--------------------------------|----------------------------------------|
+| Content Retrieval | Full page or summary           | Semantic chunks                        |
+| Relevance         | Keyword-based                  | Semantic similarity                    |
+| Token Efficiency  | May include irrelevant content | Only relevant chunks                   |
+| Processing        | Simpler, faster                | More intelligent, slower               |
+| Cost              | Lower                          | Higher (RAG processing)                |
+| Best For          | General search                 | Deep research, technical queries       |
+
+---
+
+## Deprecated: web_search_albert_rag
+
+### ⚠️ Deprecation Notice
+
+The `web_search_albert_rag` tool is **deprecated** and should not be used in new implementations.
+
+**Replacement**: Use `web_search_brave_with_document_backend` instead, which provides:
+- Better performance
+- More control over the RAG backend
+- Temporary collections (no cleanup issues)
+- Token usage tracking
+- Parallel processing support
+
+### Why Deprecated?
+
+- Limited to Albert API only
+- No control over document backend
+- Less flexible than the new approach
+- Maintenance burden
+
+### Timeline
+
+- **Current**: Still functional but not recommended
+- **Future**: Will be removed in a future version
+
+---
+
+## Comparison
+
+### When to Use Which Tool?
+
+#### Use `web_search_brave`
+
+✅ **Best for**:
+- General web search queries
+- Quick information retrieval
+- When speed is important
+- Lower cost requirements
+- Simple fact-finding
+
+❌ **Not ideal for**:
+- Deep research requiring precise context
+- Technical documentation queries
+- When semantic relevance is crucial
+
+#### Use `web_search_brave_with_document_backend`
+
+✅ **Best for**:
+- Complex technical queries
+- Research requiring precise context
+- When semantic relevance is important
+- Questions needing deep understanding
+- Documentation and how-to queries
+
+❌ **Not ideal for**:
+- Simple factual queries
+- When speed is critical
+- Budget-constrained scenarios
+- High-volume usage
+
+---
+
+## Best Practices
+
+### Query Formulation
+
+Help the LLM formulate effective queries:
+
+```python
+# Good queries
+"Python asyncio tutorial 2024"
+"Django REST framework authentication"
+"React hooks best practices"
+
+# Poor queries
+"tell me about programming"  # Too vague
+"how do I do the thing with the stuff"  # Unclear
+```
+
+### Performance Optimization
+
+#### 1. Optimize Cache
+
+```bash
+# Longer cache for stable content
+BRAVE_CACHE_TTL=3600  # 1 hour
+
+# Shorter cache for dynamic content
+BRAVE_CACHE_TTL=300   # 5 minutes
+```
+
+#### 2. Control Result Count
+
+```bash
+# Fewer results = faster responses
+BRAVE_MAX_RESULTS=5
+
+# More results = more comprehensive
+BRAVE_MAX_RESULTS=10
+```
+
+### Summarization Best Practices
+
+Only enable summarization when needed:
+
+```bash
+# Enable for long-form content
+BRAVE_SUMMARIZATION_ENABLED=True
+
+# Disable for speed
+BRAVE_SUMMARIZATION_ENABLED=False
+```
+
+**Cost consideration**: Summarization makes additional LLM calls for each result, 
+significantly increasing costs (and execution time).
+
+### RAG Configuration
+
+For `web_search_brave_with_document_backend`:
+
+```bash
+# More chunks = more context, higher cost
+BRAVE_RAG_WEB_SEARCH_CHUNK_NUMBER=10
+
+# Fewer chunks = faster, less context
+BRAVE_RAG_WEB_SEARCH_CHUNK_NUMBER=5
+```
+
+### Search Parameters
+
+```bash
+# Localize results
+BRAVE_SEARCH_COUNTRY=FR
+BRAVE_SEARCH_LANG=fr
+
+# Safe search for public deployments
+BRAVE_SEARCH_SAFE_SEARCH=strict
+
+# Enable spell check for better results
+BRAVE_SEARCH_SPELLCHECK=True
+```
+
+---
+
+## Troubleshooting
+
+### Common Issues
+
+#### 1. No Results Returned
+
+**Symptoms**: Empty results or no snippets
+
+**Causes**:
+- Query too specific
+- Content extraction failed
+- Trafilatura couldn't parse the pages
+
+**Solutions**:
+```bash
+# Enable extra snippets
+BRAVE_SEARCH_EXTRA_SNIPPETS=True
+
+# Increase result count
+BRAVE_MAX_RESULTS=10
+
+# Check logs for extraction errors
+```
+
+#### 2. API Errors
+
+**Symptoms**: HTTP errors, authentication failures
+
+**Causes**:
+- Invalid API key
+- Rate limit exceeded
+- API service issues
+
+**Solutions**:
+```bash
+# Verify API key is set
+echo $BRAVE_API_KEY
+
+# Check Brave API dashboard for limits
+# Implement rate limiting in your application
+```
+
+#### 3. The tool is not being called
+**Symptoms**: LLM doesn't use the tool even when appropriate
+
+**Causes**:
+- Web search not enabled for the conversation
+- Tool not in model configuration
+
+**Solutions**:
+- Check conversation settings have `web_search_enabled=True`
+- Verify tool is in the model's `tools` list
+
+---
+
+## Security Considerations
+
+This tool is quite "raw", so be cautious about:
+- the results returned by the web search
+- the context size which might be large when not using summarization or RAG if long results are returned
+- the query content which might include sensitive information
+- ...
+
+### Content Validation
+
+Be aware that fetched content may contain:
+- Malicious scripts (mitigated by text extraction)
+- Inappropriate content
+- Misinformation
+- Biased information
+
+The LLM should evaluate sources critically.
+
+
+---
+
+## See Also
+
+- [Tools Overview](../tools.md)
+- [Tavily Web Search Tool](web_search_tavily.md)
+- [LLM Configuration](../llm-configuration.md)
+- [Environment Variables](../env.md)
+- [Brave Search API Documentation](https://brave.com/search/api/)
+

docs/tools/web_search_tavily.md

@@ -0,0 +1,370 @@
+# web_search_tavily Tool
+
+## Overview
+
+The `web_search_tavily` tool enables the conversation agent to search the web for up-to-date 
+information using the [Tavily Search API](https://tavily.com/).
+
+## Purpose
+
+This tool allows the LLM to:
+- Access current, real-time information beyond its training data
+- Answer questions about recent events, news, or developments
+- Provide factual information with sources
+- Retrieve specific information from the web
+
+## Configuration
+
+### Prerequisites
+
+1. **Tavily API Key**: Sign up at [Tavily](https://tavily.com/) to get an API key
+2. **Environment Variables**: Configure the required settings
+
+### Environment Variables
+
+| Variable             | Required | Default | Description                                |
+|----------------------|----------|---------|--------------------------------------------|
+| `TAVILY_API_KEY`     | **Yes**  | None    | Your Tavily API key                        |
+| `TAVILY_MAX_RESULTS` | No       | 5       | Maximum number of search results to return |
+| `TAVILY_API_TIMEOUT` | No       | 10      | API request timeout in seconds             |
+
+### Configuration Example
+
+```bash
+# .env file
+TAVILY_API_KEY=tvly-your-api-key-here
+TAVILY_MAX_RESULTS=5
+TAVILY_API_TIMEOUT=10
+```
+
+### Add to Model
+
+To enable this tool for a model, add it to the `tools` list in your LLM configuration:
+
+```json
+{
+  "models": [
+    {
+      "hrid": "my-model",
+      "tools": [
+        "web_search_tavily"
+      ]
+    }
+  ]
+}
+```
+
+Or via environment variable when using local environment settings:
+
+```ini
+AI_AGENT_TOOLS=web_search_tavily
+```
+
+## Function Signature
+
+```python
+def web_search_tavily(query: str) -> list[dict]:
+    """
+    Search the web for up-to-date information
+
+    Args:
+        query (str): The query to search for.
+
+    Returns:
+        list[dict]: A list of search results, each represented as a dictionary.
+    """
+```
+
+## Parameters
+
+| Parameter | Type | Required | Description             |
+|-----------|------|----------|-------------------------|
+| `query`   | str  | Yes      | The search query string |
+
+## Return Value
+
+Returns a list of dictionaries, each containing:
+
+```python
+{
+    "link": str,      # URL of the result
+    "title": str,     # Title of the page
+    "snippet": str    # Content snippet from the page
+}
+```
+
+### Example Return Value
+
+```python
+[
+    {
+        "link": "https://example.com/article1",
+        "title": "Introduction to Python",
+        "snippet": "Python is a high-level programming language known for its simplicity..."
+    },
+    {
+        "link": "https://example.com/article2",
+        "title": "Python Best Practices",
+        "snippet": "Follow these best practices to write clean and efficient Python code..."
+    }
+]
+```
+
+## How the LLM Uses It
+
+When a user asks for current information or specific facts:
+
+1. **LLM recognizes** the need for external information
+2. **Formulates** an appropriate search query
+3. **Calls** `web_search_tavily(query="search terms")`
+4. **Receives** a list of search results
+5. **Synthesizes** the information into a response
+6. **Provides** the answer with source references
+
+### Example Conversation
+
+**User**: "What are the latest developments in quantum computing?"
+
+**LLM** (internal): *Calls `web_search_tavily("latest developments quantum computing 2024")`*
+
+**Tool Response**: 
+```python
+[
+    {
+        "link": "https://techcrunch.com/quantum-news",
+        "title": "Major Breakthrough in Quantum Computing",
+        "snippet": "Researchers announced a significant breakthrough..."
+    },
+    # ... more results
+]
+```
+
+**LLM** (to user): "Based on recent sources, there have been several developments in quantum computing. 
+Researchers recently announced a breakthrough in error correction. Additionally, new quantum processors 
+with improved qubit stability have been unveiled..."
+
+## Implementation Details
+
+### Source Code
+
+Located at: `src/backend/chat/tools/web_search_tavily.py`
+
+```python
+"""Web search tool using Tavily for the chat agent."""
+
+from django.conf import settings
+
+import requests
+
+
+def web_search_tavily(query: str) -> list[dict]:
+    """
+    Search the web for up-to-date information
+
+    Args:
+        query (str): The query to search for.
+
+    Returns:
+        list[dict]: A list of search results, each represented as a dictionary.
+    """
+    url = "https://api.tavily.com/search"
+    data = {
+        "query": query,
+        "api_key": settings.TAVILY_API_KEY,
+        "max_results": settings.TAVILY_MAX_RESULTS,
+    }
+    response = requests.post(url, json=data, timeout=settings.TAVILY_API_TIMEOUT)
+    response.raise_for_status()
+
+    json_response = response.json()
+
+    raw_search_results = json_response.get("results", [])
+
+    return [
+        {
+            "link": result["url"],
+            "title": result.get("title", ""),
+            "snippet": result.get("content"),
+        }
+        for result in raw_search_results
+    ]
+```
+
+### Registration
+
+The tool is registered in `src/backend/chat/tools/__init__.py`:
+
+```python
+"web_search_tavily": Tool(
+    web_search_tavily, 
+    takes_ctx=False, 
+    prepare=only_if_web_search_enabled
+)
+```
+
+Note that:
+- `takes_ctx=False` - This tool doesn't need the conversation context
+- `prepare=only_if_web_search_enabled` - Only available when web search is enabled
+
+## Django Settings
+
+The tool uses these Django settings from `settings.py`:
+
+```python
+# Tavily API
+TAVILY_API_KEY = values.Value(
+    None,  # Tavily API key is not set by default
+    environ_name="TAVILY_API_KEY",
+    environ_prefix=None,
+)
+TAVILY_MAX_RESULTS = values.PositiveIntegerValue(
+    default=5,
+    environ_name="TAVILY_MAX_RESULTS",
+    environ_prefix=None,
+)
+TAVILY_API_TIMEOUT = values.PositiveIntegerValue(
+    default=10,  # seconds
+    environ_name="TAVILY_API_TIMEOUT",
+    environ_prefix=None,
+)
+```
+
+## Error Handling
+
+The tool may raise exceptions in the following cases:
+
+### Missing API Key
+```python
+# If TAVILY_API_KEY is not set
+AttributeError: 'Settings' object has no attribute 'TAVILY_API_KEY'
+```
+
+**Solution**: Set the `TAVILY_API_KEY` environment variable
+
+### API Errors
+```python
+# If the API request fails
+requests.exceptions.HTTPError: 401 Unauthorized
+```
+
+**Possible causes**:
+- Invalid API key
+- Exceeded rate limits
+- API service unavailable
+
+### Timeout Errors
+```python
+# If the request takes too long
+requests.exceptions.Timeout
+```
+
+**Solution**: Increase `TAVILY_API_TIMEOUT` or check network connectivity
+
+## Best Practices
+
+### Query Formulation
+
+The LLM should formulate queries that are:
+- **Specific and focused** - Better results with targeted queries
+- **Up-to-date** - Include year or "latest" when relevant
+- **Clear** - Avoid ambiguous terms
+- **Concise** - Remove unnecessary words
+
+Good query examples:
+- ✅ "quantum computing breakthroughs 2024"
+- ✅ "latest Python 3.12 features"
+- ✅ "climate change COP29 outcomes"
+
+Poor query examples:
+- ❌ "tell me about stuff happening" (too vague)
+- ❌ "what is the weather like today in Paris on November 5th 2024 at 3pm" (too specific/long)
+
+### Rate Limiting
+
+Be aware of Tavily API rate limits:
+- Free tier: Limited requests per month
+- Paid tiers: Higher limits
+
+Monitor your usage and implement caching if needed.
+
+### Result Count
+
+The `TAVILY_MAX_RESULTS` setting controls how many results are returned:
+- **Lower values (3-5)**: Faster responses, less context for LLM
+- **Higher values (8-10)**: More comprehensive, but slower and more expensive
+
+Recommended: **5 results** for most use cases
+
+## Troubleshooting
+
+### Tool Not Being Called
+
+**Symptoms**: LLM doesn't use web search even when appropriate
+
+**Possible causes**:
+1. Web search not enabled for the conversation
+2. Tool not in model configuration
+3. API key not set
+
+**Solutions**:
+1. Check conversation settings have `web_search_enabled=True`
+2. Verify tool is in the model's `tools` list
+3. Confirm `TAVILY_API_KEY` is set
+
+### No Results Returned
+
+**Symptoms**: Tool returns empty list
+
+**Possible causes**:
+1. Query too specific
+2. No matching results
+3. API filtering results
+
+**Solutions**:
+1. Try broader query terms
+2. Check Tavily dashboard for query logs
+3. Review API response in logs
+
+### Slow Responses
+
+**Symptoms**: Tool takes a long time to respond
+
+**Possible causes**:
+1. Network latency
+2. Tavily API slow
+3. Timeout too high
+
+**Solutions**:
+1. Check network connectivity
+2. Monitor Tavily status page
+3. Adjust `TAVILY_API_TIMEOUT` if needed
+
+## Security Considerations
+
+This tool is quite "raw", and was currently only used for test purpose, so be cautious about:
+- the results returned by the web search
+- the context size which might be large if many results are returned
+- the query content which might include sensitive information
+- ...
+
+## Performance Optimization
+
+### Query Optimization
+
+You may want to help the LLM formulate better queries by including something like this in the system prompt:
+
+```
+When using web search:
+- Use specific, focused queries
+- Include relevant time periods if needed
+- Avoid unnecessary words
+- Combine related terms
+```
+
+## See Also
+
+- [Tools Overview](../tools.md)
+- [Brave Web Search Tool](web_search_brave.md)
+- [Web Search Configuration](../llm-configuration.md)
+- [Environment Variables](../env.md)
+

env.d/development/common.dist

@@ -53,5 +53,3 @@ OIDC_AUTH_REQUEST_EXTRA_PARAMS={"acr_values": "eidas1"}
 # AI_BASE_URL=https://openaiendpoint.com
 AI_API_KEY=password
 # AI_MODEL=llama
-
-ML_FLOW_TRACKING_URI = "http://ml-flow:5050"

env.d/development/common.e2e.dist

@@ -1,4 +1,12 @@
 # For the CI job test-e2e
 BURST_THROTTLE_RATES="200/minute"
-DJANGO_SERVER_TO_SERVER_API_TOKENS=test-e2e
 SUSTAINED_THROTTLE_RATES="200/hour"
+
+# LLM
+LLM_CONFIGURATION_FILE_PATH = /app/conversations/configuration/llm/default.e2e.json
+
+# Features
+FEATURE_FLAG_WEB_SEARCH=ENABLED
+FEATURE_FLAG_DOCUMENT_UPLOAD=ENABLED
+
+AUTO_TITLE_AFTER_USER_MESSAGES=3

package.json

@@ -1,6 +0,0 @@
-{
-  "dependencies": {
-    "@ai-sdk/react": "^1.2.12",
-    "@ai-sdk/ui-utils": "^1.2.11"
-  }
-}

renovate.json

@@ -15,6 +15,24 @@
       "matchPackageNames": ["redis"],
       "allowedVersions": "<6.0.0"
     },
+    {
+      "groupName": "ignore recent markitdown versions",
+      "matchManagers": ["pep621"],
+      "matchPackageNames": ["markitdown"],
+      "allowedVersions": "==0.0.2"
+    },
+    {
+      "groupName": "ignore recent lxml versions not supported by htmldate==1.9.3",
+      "matchManagers": ["pep621"],
+      "matchPackageNames": ["lxml"],
+      "allowedVersions": "<6"
+    },
+    {
+      "groupName": "ignore recent pylint versions not supported by pylint-django",
+      "matchManagers": ["pep621"],
+      "matchPackageNames": ["pylint"],
+      "allowedVersions": "<4"
+    },
     {
       "enabled": false,
       "groupName": "ignored js dependencies",
@@ -26,6 +44,12 @@
         "node-fetch",
         "workbox-webpack-plugin"
       ]
+    },
+    {
+      "groupName": "ignore Vercel SDK >= 5.0.0",
+      "matchManagers": ["npm"],
+      "matchPackageNames": ["@ai-sdk/react", "@ai-sdk/ui-utils"],
+      "allowedVersions": "^1.2.0"
     }
   ]
 }

src/OpenMockLLM/Dockerfile

@@ -0,0 +1,19 @@
+FROM python:3.13.3-alpine
+
+# Upgrade pip to its latest release to speed up dependencies installation
+RUN python -m pip install --upgrade pip setuptools lorem-text
+
+# Upgrade system packages to install security updates
+RUN apk update && \
+  apk upgrade
+
+RUN apk add --no-cache git
+
+# Install the package
+RUN pip install git+https://github.com/etalab-ia/openmockllm.git
+
+# Expose the default port
+EXPOSE 8000
+
+# Set default command
+CMD ["openmockllm", "--host", "0.0.0.0", "--port", "8000"]

src/OpenMockLLM/README.md

@@ -0,0 +1,19 @@
+[OpenMockLLM](https://github.com/etalab-ia/OpenMockLLM) is a FastAPI-based mock LLM API server that simulates 
+several Large Language Model API providers.
+
+This is a simple docker image to run the server for testing and development purposes (E2E tests mainly).
+
+It's a bit overkill to have a dedicated image for that, but it allows simple E2E stack with docker-compose since
+our code is also run in Docker containers.
+
+## Build and Run manually
+
+```bash
+docker build -t openmockllm .
+docker run -p 8000:8000 openmockllm
+```
+
+## Next steps
+
+- Add more chat completion behaviors (specific text streaming, function calling, etc.)
+- Pin a specific OpenMockLLM version in the Dockerfile

src/backend/.pylintrc

@@ -23,7 +23,9 @@ jobs=0
 
 # List of plugins (as comma separated values of python modules names) to load,
 # usually to register additional checkers.
-load-plugins=pylint_django,pylint.extensions.no_self_use
+load-plugins=pylint_django,
+             pylint.extensions.no_self_use,
+             pylint_pydantic,
 
 # Pickle collected data for later comparisons.
 persistent=yes

src/backend/activation_codes/admin.py

@@ -0,0 +1,349 @@
+"""Admin classes for activation codes application."""
+
+from django.conf import settings
+from django.contrib import admin
+from django.utils.html import format_html, format_html_join
+from django.utils.translation import gettext_lazy as _
+
+from . import models
+
+
+@admin.register(models.ActivationCode)
+class ActivationCodeAdmin(admin.ModelAdmin):
+    """Admin class for ActivationCode model"""
+
+    list_display = (
+        "code",
+        "usage_display",
+        "is_active",
+        "expires_at",
+        "created_at",
+        "description_short",
+    )
+
+    list_filter = (
+        "is_active",
+        "created_at",
+        "expires_at",
+    )
+
+    search_fields = (
+        "code",
+        "description",
+    )
+
+    readonly_fields = (
+        "id",
+        "current_uses",
+        "created_at",
+        "updated_at",
+        "usage_details",
+    )
+
+    fieldsets = (
+        (
+            None,
+            {
+                "fields": (
+                    "id",
+                    "code",
+                    "description",
+                )
+            },
+        ),
+        (
+            _("Configuration"),
+            {
+                "fields": (
+                    "max_uses",
+                    "current_uses",
+                    "is_active",
+                    "expires_at",
+                )
+            },
+        ),
+        (
+            _("Usage details"),
+            {"fields": ("usage_details",)},
+        ),
+        (
+            _("Timestamps"),
+            {
+                "fields": (
+                    "created_at",
+                    "updated_at",
+                )
+            },
+        ),
+    )
+
+    actions = ["recompute_current_uses"]
+
+    def get_readonly_fields(self, request, obj=None):
+        """Make `code` readonly when editing an existing ActivationCode.
+
+        When obj is None (creation form), `code` remains editable. When obj is
+        provided (editing), add `code` to readonly fields so it cannot be
+        changed after creation.
+        """
+        # Start from the configured readonly_fields to preserve other read-only fields
+        ro_fields = list(self.readonly_fields)
+        if obj is not None:
+            ro_fields.append("code")
+        return tuple(ro_fields)
+
+    def usage_display(self, obj):
+        """Display usage statistics."""
+        max_uses = obj.max_uses if obj.max_uses > 0 else "∞"
+        if obj.current_uses >= obj.max_uses and obj.max_uses > 0:
+            color = "red"
+        elif obj.current_uses > 0:
+            color = "orange"
+        else:
+            color = "green"
+
+        return format_html(
+            '<span style="color: {};">{} / {}</span>', color, obj.current_uses, max_uses
+        )
+
+    usage_display.short_description = _("Usage")
+
+    def description_short(self, obj):
+        """Display truncated description."""
+        if obj.description:
+            return obj.description[:50] + "..." if len(obj.description) > 50 else obj.description
+        return "-"
+
+    description_short.short_description = _("Description")
+
+    def usage_details(self, obj):
+        """Display detailed usage information."""
+        usages = obj.usages.select_related("user").all()
+
+        if not usages:
+            return _("No users have used this code yet")
+
+        table_head = format_html(
+            (
+                "<table style='width: 100%; border-collapse: collapse;'>"
+                "<tr style='background-color: #f0f0f0;'>"
+                "<th style='padding: 8px; text-align: left;'>{name}</th>"
+                "<th style='padding: 8px; text-align: left;'>{title}</th>"
+                "<th style='padding: 8px; text-align: left;'>{date}</th>"
+                "</tr>"
+            ),
+            name=_("Name"),
+            title=_("Email"),
+            date=_("Date"),
+        )
+
+        rows = format_html_join(
+            "",
+            (
+                "<tr style='border-bottom: 1px solid #ddd;'>"
+                "<td style='padding: 8px;'>{name}</td>"
+                "<td style='padding: 8px;'>{email}</td>"
+                "<td style='padding: 8px;'>{created_at}</td>"
+                "</tr>"
+            ),
+            (
+                {
+                    "name": usage.user.full_name or "-",
+                    "email": usage.user.email or "-",
+                    "created_at": usage.created_at.strftime("%Y-%m-%d %H:%M"),
+                }
+                for usage in usages
+            ),
+        )
+
+        return format_html("{table_head}{rows}</table>", table_head=table_head, rows=rows)
+
+    usage_details.short_description = _("Users who used this code")
+
+    @admin.action(description=_("Recompute current uses from related activations"))
+    def recompute_current_uses(self, request, queryset):
+        """Recompute the current_uses field by counting related UserActivation objects."""
+        updated_count = 0
+        for activation_code in queryset:
+            actual_uses = activation_code.usages.count()
+            if activation_code.current_uses != actual_uses:
+                activation_code.current_uses = actual_uses
+                activation_code.save(update_fields=["current_uses", "updated_at"])
+                updated_count += 1
+
+        if updated_count == 0:
+            self.message_user(
+                request,
+                _("All selected activation codes already have correct usage counts."),
+            )
+        else:
+            self.message_user(
+                request,
+                _("Successfully recomputed usage counts for %(count)d activation code(s).")
+                % {"count": updated_count},
+            )
+
+
+@admin.register(models.UserActivation)
+class UserActivationAdmin(admin.ModelAdmin):
+    """Admin class for UserActivation model"""
+
+    list_display = (
+        "user_display",
+        "user_email",
+        "activation_code",
+        "created_at",
+    )
+
+    list_filter = ("created_at",)
+
+    search_fields = (
+        "user__email",
+        "user__full_name",
+        "activation_code__code",
+    )
+
+    readonly_fields = (
+        "id",
+        "user",
+        "activation_code",
+        "created_at",
+        "updated_at",
+    )
+
+    fieldsets = (
+        (
+            None,
+            {
+                "fields": (
+                    "id",
+                    "user",
+                    "activation_code",
+                )
+            },
+        ),
+        (
+            _("Timestamps"),
+            {
+                "fields": (
+                    "created_at",
+                    "updated_at",
+                )
+            },
+        ),
+    )
+
+    def user_display(self, obj):
+        """Display user's full name."""
+        return obj.user.full_name or str(obj.user.id)
+
+    user_display.short_description = _("User")
+
+    def user_email(self, obj):
+        """Display user's email."""
+        return obj.user.email or "-"
+
+    user_email.short_description = _("Email")
+
+    def has_add_permission(self, request):
+        """Disable manual creation of user activations."""
+        return False
+
+
+@admin.register(models.UserRegistrationRequest)
+class UserRegistrationRequestAdmin(admin.ModelAdmin):
+    """Admin class for UserRegistrationRequest model"""
+
+    list_display = (
+        "user_display",
+        "created_at",
+        "has_user_activation",
+    )
+
+    readonly_fields = (
+        "id",
+        "user",
+        "created_at",
+        "updated_at",
+        "user_activation",
+    )
+
+    search_fields = (
+        "user__email",
+        "user__full_name",
+    )
+
+    list_filter = ("created_at",)
+
+    actions = ["add_to_brevo_waiting_list", "remove_from_brevo_waiting_list"]
+
+    def user_display(self, obj):
+        """Display user's full name."""
+        return obj.user.email or str(obj.user.pk)
+
+    user_display.short_description = _("User")
+
+    def has_user_activation(self, obj):
+        """Indicate if the user has used an activation code."""
+        return obj.user_activation_id is not None
+
+    has_user_activation.boolean = True
+    has_user_activation.short_description = _("Has used activation code")
+
+    @admin.action(description=_("Add selected users to Brevo waiting list"))
+    def add_to_brevo_waiting_list(self, request, queryset):
+        """Add selected users to Brevo waiting list."""
+        # pylint: disable=import-outside-toplevel
+        from core.brevo import add_user_to_brevo_list  # noqa: PLC0415
+
+        registration_to_send = queryset.filter(
+            user_activation__isnull=True,
+        )
+
+        _total_emails = 0
+        for i in range(0, registration_to_send.count(), 150):
+            batch = registration_to_send[i : i + 150]
+            emails = [reg.user.email for reg in batch if reg.user.email]
+            if emails:
+                add_user_to_brevo_list(emails, settings.BREVO_WAITING_LIST_ID)
+                _total_emails += len(emails)
+
+        if _total_emails:
+            self.message_user(
+                request,
+                _("Added %(count)d user(s) to Brevo waiting list.") % {"count": _total_emails},
+            )
+        else:
+            self.message_user(
+                request,
+                _("No valid email address found in selected registrations."),
+                level="warning",
+            )
+
+    @admin.action(description=_("Remove selected users from Brevo waiting list"))
+    def remove_from_brevo_waiting_list(self, request, queryset):
+        """Remove selected users from Brevo waiting list."""
+        # pylint: disable=import-outside-toplevel
+        from core.brevo import remove_user_from_brevo_list  # noqa: PLC0415
+
+        registration_to_send = queryset.filter(
+            user_activation__isnull=False,
+        )
+        _total_emails = 0
+        for i in range(0, registration_to_send.count(), 150):
+            batch = registration_to_send[i : i + 150]
+            emails = [reg.user.email for reg in batch if reg.user.email]
+            if emails:
+                remove_user_from_brevo_list(emails, settings.BREVO_WAITING_LIST_ID)
+                _total_emails += len(emails)
+        if _total_emails:
+            self.message_user(
+                request,
+                _("Removed %(count)d user(s) from Brevo waiting list.") % {"count": _total_emails},
+            )
+        else:
+            self.message_user(
+                request,
+                _("No valid email address found in selected registrations."),
+                level="warning",
+            )

src/backend/activation_codes/exceptions.py

@@ -0,0 +1,9 @@
+"""Exceptions for activation code handling."""
+
+
+class InvalidCodeError(ValueError):
+    """Raised when an activation code is invalid or cannot be used."""
+
+
+class UserAlreadyActivatedError(ValueError):
+    """Raised when a user tries to activate but is already activated."""

src/backend/activation_codes/factories.py

@@ -0,0 +1,29 @@
+"""Factories for creating activation code and user activation instances for testing."""
+
+from django.utils import timezone
+
+import factory.django
+
+from core.factories import UserFactory
+
+from . import models
+
+
+class ActivationCodeFactory(factory.django.DjangoModelFactory):
+    """A factory to create activation codes for testing purposes."""
+
+    class Meta:
+        model = models.ActivationCode
+
+    code = factory.LazyAttribute(lambda x: models.generate_activation_code())
+    created_at = factory.LazyAttribute(lambda obj: timezone.now())
+
+
+class UserActivationFactory(factory.django.DjangoModelFactory):
+    """A factory to create user activations for testing purposes."""
+
+    class Meta:
+        model = models.UserActivation
+
+    user = factory.SubFactory(UserFactory)
+    activation_code = factory.SubFactory(ActivationCodeFactory)

src/backend/activation_codes/migrations/0001_initial.py

@@ -0,0 +1,234 @@
+# Generated by Django 5.2.7 on 2025-10-09 08:30
+
+import uuid
+
+import django.db.models.deletion
+from django.conf import settings
+from django.db import migrations, models
+
+import activation_codes.models
+
+
+class Migration(migrations.Migration):
+    initial = True
+
+    dependencies = [
+        migrations.swappable_dependency(settings.AUTH_USER_MODEL),
+    ]
+
+    operations = [
+        migrations.CreateModel(
+            name="ActivationCode",
+            fields=[
+                (
+                    "id",
+                    models.UUIDField(
+                        default=uuid.uuid4,
+                        editable=False,
+                        help_text="primary key for the record as UUID",
+                        primary_key=True,
+                        serialize=False,
+                        verbose_name="id",
+                    ),
+                ),
+                (
+                    "created_at",
+                    models.DateTimeField(
+                        auto_now_add=True,
+                        help_text="date and time at which a record was created",
+                        verbose_name="created on",
+                    ),
+                ),
+                (
+                    "updated_at",
+                    models.DateTimeField(
+                        auto_now=True,
+                        help_text="date and time at which a record was last updated",
+                        verbose_name="updated on",
+                    ),
+                ),
+                (
+                    "code",
+                    models.CharField(
+                        default=activation_codes.models.generate_activation_code,
+                        help_text="The activation code that users will enter",
+                        max_length=50,
+                        unique=True,
+                        validators=[
+                            django.core.validators.RegexValidator(
+                                message="Code must be alphanumeric and contain no spaces or special characters",
+                                regex="^[A-Z0-9]+$",
+                            )
+                        ],
+                        verbose_name="activation code",
+                    ),
+                ),
+                (
+                    "max_uses",
+                    models.PositiveIntegerField(
+                        default=1,
+                        help_text="Maximum number of times this code can be used. 0 means unlimited.",
+                        verbose_name="maximum uses",
+                    ),
+                ),
+                (
+                    "current_uses",
+                    models.PositiveIntegerField(
+                        default=0,
+                        editable=False,
+                        help_text="Number of times this code has been used",
+                        verbose_name="current uses",
+                    ),
+                ),
+                (
+                    "is_active",
+                    models.BooleanField(
+                        default=True,
+                        help_text="Whether this code can still be used",
+                        verbose_name="active",
+                    ),
+                ),
+                (
+                    "expires_at",
+                    models.DateTimeField(
+                        blank=True,
+                        help_text="Date and time when this code expires",
+                        null=True,
+                        verbose_name="expires at",
+                    ),
+                ),
+                (
+                    "description",
+                    models.TextField(
+                        blank=True,
+                        help_text="Internal description or notes about this code",
+                        verbose_name="description",
+                    ),
+                ),
+            ],
+            options={
+                "verbose_name": "activation code",
+                "verbose_name_plural": "activation codes",
+                "db_table": "activation_code",
+                "ordering": ["-created_at"],
+            },
+        ),
+        migrations.CreateModel(
+            name="UserActivation",
+            fields=[
+                (
+                    "id",
+                    models.UUIDField(
+                        default=uuid.uuid4,
+                        editable=False,
+                        help_text="primary key for the record as UUID",
+                        primary_key=True,
+                        serialize=False,
+                        verbose_name="id",
+                    ),
+                ),
+                (
+                    "created_at",
+                    models.DateTimeField(
+                        auto_now_add=True,
+                        help_text="date and time at which a record was created",
+                        verbose_name="created on",
+                    ),
+                ),
+                (
+                    "updated_at",
+                    models.DateTimeField(
+                        auto_now=True,
+                        help_text="date and time at which a record was last updated",
+                        verbose_name="updated on",
+                    ),
+                ),
+                (
+                    "activation_code",
+                    models.ForeignKey(
+                        help_text="The activation code that was used",
+                        on_delete=django.db.models.deletion.PROTECT,
+                        related_name="usages",
+                        to="activation_codes.activationcode",
+                        verbose_name="activation code",
+                    ),
+                ),
+                (
+                    "user",
+                    models.OneToOneField(
+                        help_text="The user who used the activation code",
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="activation",
+                        to=settings.AUTH_USER_MODEL,
+                        verbose_name="user",
+                    ),
+                ),
+            ],
+            options={
+                "verbose_name": "user activation",
+                "verbose_name_plural": "user activations",
+                "db_table": "user_activation",
+                "ordering": ["-created_at"],
+            },
+        ),
+        migrations.CreateModel(
+            name="UserRegistrationRequest",
+            fields=[
+                (
+                    "id",
+                    models.UUIDField(
+                        default=uuid.uuid4,
+                        editable=False,
+                        help_text="primary key for the record as UUID",
+                        primary_key=True,
+                        serialize=False,
+                        verbose_name="id",
+                    ),
+                ),
+                (
+                    "created_at",
+                    models.DateTimeField(
+                        auto_now_add=True,
+                        help_text="date and time at which a record was created",
+                        verbose_name="created on",
+                    ),
+                ),
+                (
+                    "updated_at",
+                    models.DateTimeField(
+                        auto_now=True,
+                        help_text="date and time at which a record was last updated",
+                        verbose_name="updated on",
+                    ),
+                ),
+                (
+                    "user",
+                    models.OneToOneField(
+                        help_text="The user who made the registration request",
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="registration_request",
+                        to=settings.AUTH_USER_MODEL,
+                        verbose_name="user",
+                    ),
+                ),
+                (
+                    "user_activation",
+                    models.OneToOneField(
+                        blank=True,
+                        help_text="Store if the user received an activation code and used it",
+                        null=True,
+                        on_delete=django.db.models.deletion.SET_NULL,
+                        related_name="registration_request",
+                        to="activation_codes.useractivation",
+                        verbose_name="user activation",
+                    ),
+                ),
+            ],
+            options={
+                "verbose_name": "user registration request",
+                "verbose_name_plural": "user registration requests",
+                "db_table": "user_registration_request",
+                "ordering": ["-created_at"],
+            },
+        ),
+    ]

src/backend/activation_codes/models.py

@@ -0,0 +1,226 @@
+"""
+Models for the activation codes application
+"""
+
+import logging
+import secrets
+import string
+
+from django.conf import settings
+from django.core.exceptions import ValidationError
+from django.core.validators import RegexValidator
+from django.db import IntegrityError, models, transaction
+from django.utils import timezone
+from django.utils.translation import gettext_lazy as _
+
+from core.brevo import add_user_to_brevo_list, remove_user_from_brevo_list
+from core.models import BaseModel, User
+
+from activation_codes.exceptions import InvalidCodeError, UserAlreadyActivatedError
+
+logger = logging.getLogger(__name__)
+
+
+def generate_activation_code():
+    """Generate a random 16-character activation code."""
+    alphabet = string.ascii_uppercase + string.digits
+    # Remove ambiguous characters
+    alphabet = alphabet.replace("O", "").replace("0", "").replace("I", "").replace("1", "")
+    return "".join(secrets.choice(alphabet) for _ in range(16))
+
+
+class ActivationCode(BaseModel):
+    """
+    Represents an activation code that can be used to activate user accounts.
+    """
+
+    code = models.CharField(
+        verbose_name=_("activation code"),
+        help_text=_("The activation code that users will enter"),
+        max_length=50,
+        unique=True,
+        default=generate_activation_code,
+        validators=[
+            RegexValidator(
+                regex=r"^[A-Z0-9]+$",
+                message=_("Code must be alphanumeric and contain no spaces or special characters"),
+            )
+        ],
+    )
+
+    max_uses = models.PositiveIntegerField(
+        verbose_name=_("maximum uses"),
+        help_text=_("Maximum number of times this code can be used. 0 means unlimited."),
+        default=1,
+    )
+
+    current_uses = models.PositiveIntegerField(
+        verbose_name=_("current uses"),
+        help_text=_("Number of times this code has been used"),
+        default=0,
+        editable=False,
+    )
+
+    is_active = models.BooleanField(
+        verbose_name=_("active"),
+        help_text=_("Whether this code can still be used"),
+        default=True,
+    )
+
+    expires_at = models.DateTimeField(
+        verbose_name=_("expires at"),
+        help_text=_("Date and time when this code expires"),
+        null=True,
+        blank=True,
+    )
+
+    description = models.TextField(
+        verbose_name=_("description"),
+        help_text=_("Internal description or notes about this code"),
+        blank=True,
+    )
+
+    class Meta:
+        db_table = "activation_code"
+        verbose_name = _("activation code")
+        verbose_name_plural = _("activation codes")
+        ordering = ["-created_at"]
+
+    def __str__(self):
+        """Return string representation of the activation code."""
+        return f"{self.code} ({self.current_uses}/{self.max_uses if self.max_uses > 0 else '∞'})"
+
+    def is_valid(self):
+        """Check if the code is still valid and can be used."""
+        if not self.is_active:
+            return False
+
+        if self.expires_at and self.expires_at < timezone.now():
+            return False
+
+        if self.max_uses > 0 and self.current_uses >= self.max_uses:
+            return False
+
+        return True
+
+    def can_be_used(self):
+        """Alias for is_valid() for better readability."""
+        return self.is_valid()
+
+    def use(self, user):
+        """
+        Mark this code as used by a user.
+
+        Args:
+            user: The User instance using this code
+
+        Returns:
+            UserActivation instance
+
+        Raises:
+            ValidationError: If the code cannot be used
+        """
+        with transaction.atomic():
+            # Lock the activation code row to prevent concurrent overuse.
+            locked_code = ActivationCode.objects.select_for_update().get(pk=self.pk)
+
+            if not locked_code.is_valid():
+                raise InvalidCodeError(_("This activation code is no longer valid"))
+
+            # Create activation record; rely on DB uniqueness for concurrent duplicate attempts.
+            try:
+                activation = UserActivation.objects.create(user=user, activation_code=locked_code)
+            except (IntegrityError, ValidationError) as exc:
+                # User already has an activation in a concurrent or prior transaction.
+                raise UserAlreadyActivatedError(
+                    _("You have already activated your account")
+                ) from exc
+
+            existing_registration = bool(
+                UserRegistrationRequest.objects.filter(user=user).update(user_activation=activation)
+            )
+            if existing_registration:
+                transaction.on_commit(
+                    lambda: remove_user_from_brevo_list(
+                        [user.email], settings.BREVO_WAITING_LIST_ID
+                    )
+                )
+
+            # Increment usage counter safely under the same lock.
+            locked_code.current_uses += 1
+            locked_code.save(update_fields=["current_uses", "updated_at"])
+
+            transaction.on_commit(
+                lambda: add_user_to_brevo_list([user.email], settings.BREVO_FOLLOWUP_LIST_ID)
+            )
+
+            if locked_code.max_uses > 0 and locked_code.current_uses >= locked_code.max_uses:
+                logger.warning("Activation code %s has reached its maximum uses", locked_code.code)
+
+            return activation
+
+
+class UserActivation(BaseModel):
+    """
+    Records with user used which activation code and when.
+    """
+
+    user = models.OneToOneField(
+        User,
+        verbose_name=_("user"),
+        help_text=_("The user who used the activation code"),
+        on_delete=models.CASCADE,
+        related_name="activation",
+    )
+
+    activation_code = models.ForeignKey(
+        ActivationCode,
+        verbose_name=_("activation code"),
+        help_text=_("The activation code that was used"),
+        on_delete=models.PROTECT,
+        related_name="usages",
+    )
+
+    class Meta:
+        db_table = "user_activation"
+        verbose_name = _("user activation")
+        verbose_name_plural = _("user activations")
+        ordering = ["-created_at"]
+
+    def __str__(self):
+        """Return string representation of the user activation."""
+        return f"{self.user} - {self.activation_code.code}"
+
+
+class UserRegistrationRequest(BaseModel):
+    """
+    Records of user registration requests.
+    """
+
+    user = models.OneToOneField(
+        User,
+        verbose_name=_("user"),
+        help_text=_("The user who made the registration request"),
+        on_delete=models.CASCADE,
+        related_name="registration_request",
+    )
+
+    user_activation = models.OneToOneField(
+        UserActivation,
+        verbose_name=_("user activation"),
+        help_text=_("Store if the user received an activation code and used it"),
+        on_delete=models.SET_NULL,
+        related_name="registration_request",
+        null=True,
+        blank=True,
+    )
+
+    class Meta:
+        db_table = "user_registration_request"
+        verbose_name = _("user registration request")
+        verbose_name_plural = _("user registration requests")
+        ordering = ["-created_at"]
+
+    def __str__(self):
+        """Return string representation of the user registration request."""
+        return f"Registration request by {self.user}"

src/backend/activation_codes/permissions.py

@@ -0,0 +1,39 @@
+"""Permission classes for activation codes."""
+
+from django.conf import settings
+
+from rest_framework import permissions
+
+from . import models
+
+
+class IsActivatedUser(permissions.BasePermission):
+    """
+    Permission class that checks if user has activated their account.
+
+    This permission is only enforced if ACTIVATION_REQUIRED is True in settings.
+    Staff users and users without authentication requirement are always allowed.
+    """
+
+    message = "activation-required"  # Custom message to indicate activation is required to frontend
+
+    def has_permission(self, request, view):
+        """Check if user has activated their account."""
+        # If activation is not required, allow access
+        if not settings.ACTIVATION_REQUIRED:
+            return True
+
+        # Staff users can always access
+        if request.user and request.user.is_staff:
+            return True
+
+        # Anonymous users are handled by other permission classes
+        if not request.user or not request.user.is_authenticated:
+            return True
+
+        # Check if user has an activation record
+        return models.UserActivation.objects.filter(user=request.user).exists()
+
+    def has_object_permission(self, request, view, obj):
+        """Check object-level permission."""
+        return self.has_permission(request, view)

src/backend/activation_codes/serializers.py

@@ -0,0 +1,50 @@
+"""Serializers for the activation codes application."""
+
+from django.utils.translation import gettext_lazy as _
+
+from rest_framework import serializers
+
+from . import models
+
+
+class ActivationCodeValidationSerializer(serializers.Serializer):  # pylint: disable=abstract-method
+    """Serializer for validating an activation code."""
+
+    code = serializers.CharField(
+        max_length=50, required=True, help_text=_("The activation code to validate")
+    )
+
+    def validate_code(self, value):
+        """Validate that the code exists and is valid."""
+        # Normalize the code (remove spaces, convert to uppercase)
+        return value.strip().upper().replace(" ", "").replace("-", "")
+
+
+class UserActivationSerializer(serializers.ModelSerializer):
+    """Serializer for user activation records."""
+
+    code = serializers.CharField(source="activation_code.code", read_only=True)
+    activated_at = serializers.DateTimeField(source="created_at", read_only=True)
+
+    class Meta:
+        model = models.UserActivation
+        fields = ["id", "code", "activated_at"]
+        read_only_fields = ["id", "code", "activated_at"]
+
+
+class ActivationStatusSerializer(serializers.Serializer):  # pylint: disable=abstract-method
+    """Serializer for activation status response."""
+
+    is_activated = serializers.BooleanField(read_only=True)
+    activation = UserActivationSerializer(read_only=True, allow_null=True)
+    requires_activation = serializers.BooleanField(read_only=True)
+
+
+class UserRegistrationRequestSerializer(serializers.ModelSerializer):
+    """Serializer for registering a user for activation notifications."""
+
+    user = serializers.HiddenField(default=serializers.CurrentUserDefault())
+
+    class Meta:
+        model = models.UserRegistrationRequest
+        fields = ["user"]

src/backend/activation_codes/tests/test_integration.py

@@ -0,0 +1,236 @@
+"""Integration tests for activation_codes application."""
+
+from datetime import timedelta
+
+from django.utils import timezone
+
+import pytest
+from rest_framework import status
+
+from core.factories import UserFactory
+
+from activation_codes.factories import ActivationCodeFactory
+from activation_codes.models import ActivationCode, UserActivation
+
+
+@pytest.mark.django_db
+def test_complete_activation_flow(api_client, settings):
+    """Test complete user activation flow from registration to usage."""
+    settings.ACTIVATION_REQUIRED = True
+
+    # Create a user (simulating registration)
+    user = UserFactory(email="newuser@example.com", password="password123")
+
+    # Create an activation code (simulating admin creating codes)
+    activation_code = ActivationCode.objects.create(
+        code="WELCOME123456789", max_uses=10, description="Welcome batch for new users"
+    )
+
+    # User logs in
+    api_client.force_authenticate(user=user)
+
+    # Step 1: Check activation status (should not be activated)
+    response = api_client.get("/api/v1.0/activation/status/")
+    assert response.status_code == status.HTTP_200_OK
+    assert response.data["is_activated"] is False
+    assert response.data["requires_activation"] is True
+
+    # Step 2: User enters activation code
+    response = api_client.post("/api/v1.0/activation/validate/", {"code": "WELCOME123456789"})
+    assert response.status_code == status.HTTP_201_CREATED
+    assert "successfully activated" in response.data["detail"]
+
+    # Step 3: Check activation status again (should now be activated)
+    response = api_client.get("/api/v1.0/activation/status/")
+    assert response.status_code == status.HTTP_200_OK
+    assert response.data["is_activated"] is True
+    assert response.data["activation"]["code"] == "WELCOME123456789"
+
+    # Step 4: Verify in database
+    assert UserActivation.objects.filter(user=user).exists()
+    activation_code.refresh_from_db()
+    assert activation_code.current_uses == 1
+
+
+@pytest.mark.django_db
+def test_activation_not_required_flow(api_client, settings):
+    """Test that when activation is not required, users can access without codes."""
+    settings.ACTIVATION_REQUIRED = False
+
+    user = UserFactory(email="freeuser@example.com", password="password123")
+
+    api_client.force_authenticate(user=user)
+
+    # Check status
+    response = api_client.get("/api/v1.0/activation/status/")
+    assert response.status_code == status.HTTP_200_OK
+    assert response.data["requires_activation"] is False
+    assert response.data["is_activated"] is False  # Not activated but not required
+
+
+@pytest.mark.django_db
+def test_multiple_users_same_code(api_client, settings):
+    """Test multiple users using the same multi-use code."""
+    settings.ACTIVATION_REQUIRED = True
+
+    # Create a multi-use code
+    code = ActivationCode.objects.create(
+        code="TEAMCODE12345678", max_uses=3, description="Team activation code"
+    )
+
+    # Create 3 users
+    users = []
+    for i in range(3):
+        user = UserFactory(email=f"teamuser{i}@example.com", password="password123")
+        users.append(user)
+
+    # Each user activates
+    for i, user in enumerate(users):
+        api_client.force_authenticate(user=user)
+        response = api_client.post("/api/v1.0/activation/validate/", {"code": "TEAMCODE12345678"})
+        assert response.status_code == status.HTTP_201_CREATED
+
+        code.refresh_from_db()
+        assert code.current_uses == i + 1
+
+    # Code should now be exhausted
+    code.refresh_from_db()
+    assert code.is_valid() is False
+
+    # Try with a 4th user (should fail)
+    user4 = UserFactory(email="teamuser4@example.com", password="password123")
+    api_client.force_authenticate(user=user4)
+    response = api_client.post("/api/v1.0/activation/validate/", {"code": "TEAMCODE12345678"})
+    assert response.status_code == status.HTTP_400_BAD_REQUEST
+
+
+@pytest.mark.django_db
+def test_code_expiration_scenario(api_client, settings):
+    """Test code expiration over time."""
+    settings.ACTIVATION_REQUIRED = True
+
+    # Create a code that expires in 1 day
+    future_time = timezone.now() + timedelta(days=1)
+    _code = ActivationCode.objects.create(code="EXPIRES123456789", expires_at=future_time)
+
+    user = UserFactory(email="timeduser@example.com", password="password123")
+    api_client.force_authenticate(user=user)
+
+    # Should work now
+    response = api_client.post("/api/v1.0/activation/validate/", {"code": "EXPIRES123456789"})
+    assert response.status_code == status.HTTP_201_CREATED
+
+
+@pytest.mark.django_db
+def test_staff_user_bypass(api_client, settings):
+    """Test that staff users bypass activation requirement."""
+    settings.ACTIVATION_REQUIRED = True
+
+    staff_user = UserFactory(email="staff@example.com", password="password123", is_staff=True)
+
+    api_client.force_authenticate(user=staff_user)
+
+    # Staff should be able to check status even without activation
+    response = api_client.get("/api/v1.0/activation/status/")
+    assert response.status_code == status.HTTP_200_OK
+
+
+@pytest.mark.django_db
+def test_user_cannot_activate_twice(api_client, settings):
+    """Test that a user cannot activate their account twice."""
+    settings.ACTIVATION_REQUIRED = True
+
+    user = UserFactory(email="onceuser@example.com", password="password123")
+
+    _code1 = ActivationCodeFactory(code="FIRST12345678901")
+    _code2 = ActivationCodeFactory(code="SECOND1234567890")
+
+    api_client.force_authenticate(user=user)
+
+    # First activation
+    response = api_client.post("/api/v1.0/activation/validate/", {"code": "FIRST12345678901"})
+    assert response.status_code == status.HTTP_201_CREATED
+
+    # Try second activation
+    response = api_client.post("/api/v1.0/activation/validate/", {"code": "SECOND1234567890"})
+    assert response.status_code == status.HTTP_400_BAD_REQUEST
+    assert response.data == {"code": "account-already-activated"}
+
+    # Verify only one activation exists
+    assert UserActivation.objects.filter(user=user).count() == 1
+
+
+@pytest.mark.django_db
+def test_code_variations_normalized(api_client, settings):
+    """Test that different code input formats are normalized correctly."""
+    settings.ACTIVATION_REQUIRED = True
+
+    code = ActivationCodeFactory(code="TESTCODE12345678")
+
+    test_cases = [
+        "testcode12345678",  # lowercase
+        "TESTCODE12345678",  # uppercase
+        "test code 1234 5678",  # with spaces
+        "TEST-CODE-1234-5678",  # with dashes
+        " test-code 1234-5678 ",  # mixed with leading/trailing spaces
+    ]
+
+    for i, code_variation in enumerate(test_cases):
+        user = UserFactory(email=f"varuser{i}@example.com", password="password123")
+
+        # Update code to allow multiple uses
+        code.max_uses = 0
+        code.save()
+
+        api_client.force_authenticate(user=user)
+        response = api_client.post("/api/v1.0/activation/validate/", {"code": code_variation})
+        assert response.status_code == status.HTTP_201_CREATED, (
+            f"Failed for variation: {code_variation}"
+        )
+
+
+@pytest.mark.django_db
+def test_inactive_code_cannot_be_used(api_client, settings):
+    """Test that inactive codes cannot be used even if valid otherwise."""
+    settings.ACTIVATION_REQUIRED = True
+
+    _code = ActivationCodeFactory(
+        code="INACTIVE123VALID",
+        is_active=False,
+        max_uses=10,
+        expires_at=timezone.now() + timedelta(days=30),
+    )
+
+    user = UserFactory(email="blockeduser@example.com", password="password123")
+
+    api_client.force_authenticate(user=user)
+    response = api_client.post("/api/v1.0/activation/validate/", {"code": "INACTIVE123VALID"})
+    assert response.status_code == status.HTTP_400_BAD_REQUEST
+
+
+@pytest.mark.django_db
+def test_concurrent_activations_multi_use_code(api_client, settings):
+    """Test that concurrent activations don't exceed max_uses."""
+    settings.ACTIVATION_REQUIRED = True
+
+    code = ActivationCodeFactory(code="CONCURRENT123456", max_uses=2)
+
+    # Create 3 users
+    users = [
+        UserFactory(email=f"concurrent{i}@example.com", password="password123") for i in range(3)
+    ]
+
+    # First two should succeed
+    for i in range(2):
+        api_client.force_authenticate(user=users[i])
+        response = api_client.post("/api/v1.0/activation/validate/", {"code": "CONCURRENT123456"})
+        assert response.status_code == status.HTTP_201_CREATED
+
+    # Third should fail
+    api_client.force_authenticate(user=users[2])
+    response = api_client.post("/api/v1.0/activation/validate/", {"code": "CONCURRENT123456"})
+    assert response.status_code == status.HTTP_400_BAD_REQUEST
+
+    # Verify only 2 activations
+    code.refresh_from_db()
+    assert code.current_uses == 2

src/backend/activation_codes/tests/test_models.py

@@ -0,0 +1,328 @@
+"""Tests for activation_codes models."""
+
+import json
+from datetime import timedelta
+
+from django.core.exceptions import ValidationError
+from django.db.models import ProtectedError
+from django.utils import timezone
+
+import pytest
+import responses
+
+from core.factories import UserFactory
+
+from activation_codes.exceptions import InvalidCodeError, UserAlreadyActivatedError
+from activation_codes.factories import ActivationCodeFactory, UserActivationFactory
+from activation_codes.models import (
+    ActivationCode,
+    UserActivation,
+    UserRegistrationRequest,
+    generate_activation_code,
+)
+
+
+@pytest.mark.django_db
+def test_generate_activation_code():
+    """Test that generate_activation_code creates a valid code."""
+    code = generate_activation_code()
+
+    assert len(code) == 16
+    assert code.isupper()
+    assert all(c.isalnum() for c in code)
+    # Check that ambiguous characters are not present
+    assert "O" not in code
+    assert "0" not in code
+    assert "I" not in code
+    assert "1" not in code
+
+
+@pytest.mark.django_db
+def test_generate_activation_code_uniqueness():
+    """Test that generated codes are unique."""
+    codes = [generate_activation_code() for _ in range(100)]
+    assert len(codes) == len(set(codes))
+
+
+@pytest.mark.django_db
+def test_activation_code_creation():
+    """Test creating an activation code."""
+    activation_code = ActivationCodeFactory(code="TEST1234ABCD5678")
+
+    assert activation_code.code == "TEST1234ABCD5678"
+    assert activation_code.max_uses == 1
+    assert activation_code.current_uses == 0
+    assert activation_code.is_active is True
+    assert activation_code.expires_at is None
+
+
+@pytest.mark.django_db
+def test_activation_code_auto_generated_code():
+    """Test that activation code is auto-generated if not provided."""
+    code = ActivationCodeFactory()
+    assert len(code.code) == 16
+    assert code.code.isupper()
+
+
+@pytest.mark.django_db
+def test_activation_code_str_representation():
+    """Test string representation of activation code."""
+    activation_code = ActivationCodeFactory(code="TEST1234ABCD5678")
+    assert str(activation_code) == "TEST1234ABCD5678 (0/1)"
+
+
+@pytest.mark.django_db
+def test_activation_code_str_representation_unlimited():
+    """Test string representation of unlimited activation code."""
+    unlimited_activation_code = ActivationCodeFactory(code="UNLIMITED123CODE", max_uses=0)
+
+    assert str(unlimited_activation_code) == "UNLIMITED123CODE (0/∞)"
+
+
+@pytest.mark.django_db
+def test_activation_code_is_valid_active():
+    """Test that an active, non-expired code is valid."""
+    activation_code = ActivationCodeFactory()
+    assert activation_code.is_valid() is True
+    assert activation_code.can_be_used() is True
+
+
+@pytest.mark.django_db
+def test_activation_code_is_valid_inactive():
+    """Test that an inactive code is not valid."""
+    inactive_activation_code = ActivationCodeFactory(is_active=False)
+    assert inactive_activation_code.is_valid() is False
+    assert inactive_activation_code.can_be_used() is False
+
+
+@pytest.mark.django_db
+def test_activation_code_is_valid_expired():
+    """Test that an expired code is not valid."""
+    expired_activation_code = ActivationCodeFactory(
+        created_at=timezone.now() - timedelta(days=10),
+        expires_at=timezone.now() - timedelta(days=1),
+    )
+    assert expired_activation_code.is_valid() is False
+    assert expired_activation_code.can_be_used() is False
+
+
+@pytest.mark.django_db
+def test_activation_code_is_valid_max_uses_reached():
+    """Test that a code with max uses reached is not valid."""
+    activation_code = ActivationCodeFactory(max_uses=1)
+    activation_code.current_uses = 1
+    activation_code.save()
+    assert activation_code.is_valid() is False
+
+
+@pytest.mark.django_db
+def test_activation_code_is_valid_unlimited_uses():
+    """Test that unlimited code is always valid regardless of current uses."""
+    unlimited_activation_code = ActivationCodeFactory(max_uses=0)
+    unlimited_activation_code.current_uses = 100
+    unlimited_activation_code.save()
+    assert unlimited_activation_code.is_valid() is True
+
+
+@pytest.mark.django_db
+def test_activation_code_use_success():
+    """Test successfully using an activation code."""
+    user = UserFactory()
+    activation_code = ActivationCodeFactory()
+    activation = activation_code.use(user)
+
+    assert isinstance(activation, UserActivation)
+    assert activation.user == user
+    assert activation.activation_code == activation_code
+
+    # Check that usage counter was incremented
+    activation_code.refresh_from_db()
+    assert activation_code.current_uses == 1
+
+
+@pytest.mark.django_db
+def test_activation_code_use_invalid_code():
+    """Test using an invalid activation code raises error."""
+    inactive_activation_code = ActivationCodeFactory(is_active=False)
+    user = UserFactory()
+    with pytest.raises(InvalidCodeError):
+        inactive_activation_code.use(user)
+
+
+@pytest.mark.django_db
+def test_activation_code_use_already_activated():
+    """Test using a code when user is already activated raises error."""
+    user = UserFactory()
+    activation_code = ActivationCodeFactory()
+
+    # First activation
+    activation_code.use(user)
+
+    # Try to activate again with a different code
+    another_code = ActivationCodeFactory(code="ANOTHER123456789")
+    with pytest.raises(UserAlreadyActivatedError):
+        another_code.use(user)
+
+
+@pytest.mark.django_db
+def test_activation_code_use_multi_use():
+    """Test using a multi-use activation code."""
+    multi_use_activation_code = ActivationCodeFactory(max_uses=4)
+    users = [UserFactory(email=f"user{i}@example.com") for i in range(3)]
+
+    for i, user in enumerate(users):
+        activation = multi_use_activation_code.use(user)
+        assert activation.user == user
+
+        multi_use_activation_code.refresh_from_db()
+        assert multi_use_activation_code.current_uses == i + 1
+
+    # Code should still be valid
+    assert multi_use_activation_code.is_valid() is True
+
+
+@pytest.mark.django_db
+def test_activation_code_use_max_uses_exceeded():
+    """Test that code cannot be used when max uses is reached."""
+    user = UserFactory()
+    activation_code = ActivationCodeFactory(max_uses=1)
+
+    # Use the code
+    activation_code.use(user)
+
+    # Try to use it again with another user
+    another_user = UserFactory(email="another@example.com")
+    with pytest.raises(InvalidCodeError):
+        activation_code.use(another_user)
+
+
+@pytest.mark.django_db
+def test_activation_code_expiration():
+    """Test that code expires correctly."""
+    future_expiry = timezone.now() + timedelta(days=1)
+    code = ActivationCodeFactory(code="FUTURE123456789", expires_at=future_expiry)
+
+    assert code.is_valid() is True
+
+    # Manually set to past
+    code.expires_at = timezone.now() - timedelta(seconds=1)
+    code.save()
+
+    assert code.is_valid() is False
+
+
+@pytest.mark.django_db
+def test_user_activation_str_representation():
+    """Test string representation of user activation."""
+    user_activation = UserActivationFactory(activation_code__code="TEST1234ABCD5678")
+
+    expected = f"{user_activation.user} - TEST1234ABCD5678"
+    assert str(user_activation) == expected
+
+
+@pytest.mark.django_db
+def test_user_activation_one_to_one_relationship():
+    """Test that a user can only have one activation."""
+    user_activation = UserActivationFactory()
+
+    # Try to create another activation for the same user
+    with pytest.raises(ValidationError):  # should be IntegrityError
+        UserActivationFactory(user=user_activation.user)
+
+
+@pytest.mark.django_db
+def test_activation_code_protect_on_delete():
+    """Test that activation code is protected from deletion when used."""
+    user_activation = UserActivationFactory()
+
+    # Try to delete the activation code
+    with pytest.raises(ProtectedError):
+        user_activation.activation_code.delete()
+
+
+@pytest.mark.django_db
+def test_user_activation_cascade_on_user_delete():
+    """Test that activation is deleted when user is deleted."""
+    activation = UserActivationFactory()
+    activation_id = activation.pk
+
+    activation.user.delete()
+
+    # Activation should be deleted
+    assert not UserActivation.objects.filter(id=activation_id).exists()
+
+
+@pytest.mark.django_db
+def test_activation_code_ordering():
+    """Test that activation codes are ordered by created_at descending."""
+    code1 = ActivationCodeFactory(code="CODE1")
+    code2 = ActivationCodeFactory(code="CODE2")
+    code3 = ActivationCodeFactory(code="CODE3")
+
+    codes = list(ActivationCode.objects.all())
+    assert codes == [code3, code2, code1]
+
+
+@pytest.mark.django_db
+def test_user_activation_ordering():
+    """Test that user activations are ordered by created_at descending."""
+    code1 = ActivationCodeFactory(code="CODE1", max_uses=3)
+    code2 = ActivationCodeFactory(code="CODE2", max_uses=3)
+
+    user1 = UserFactory(email="user1@example.com")
+    user2 = UserFactory(email="user2@example.com")
+
+    activation1 = UserActivationFactory(user=user1, activation_code=code1)
+    activation2 = UserActivationFactory(user=user2, activation_code=code2)
+
+    activations = list(UserActivation.objects.all())
+    assert activations == [activation2, activation1]
+
+
+@responses.activate
+@pytest.mark.django_db(transaction=True)
+def test_activation_code_use_success_notify_brevo(settings):
+    """Test successfully using an activation code and notify Brevo."""
+    settings.BREVO_API_KEY = "test_brevo_api_key"
+    settings.BREVO_WAITING_LIST_ID = "test_waiting_list_id"
+    settings.BREVO_FOLLOWUP_LIST_ID = "test_followup_list_name"
+
+    brevo_remove_mock = responses.post(
+        "https://api.brevo.com/v3/contacts/lists/test_waiting_list_id/contacts/remove",
+        json={"message": "Contacts added successfully"},
+        status=201,
+    )
+
+    brevo_create_contact = responses.post(
+        "https://api.brevo.com/v3/contacts",
+        status=200,
+    )
+
+    brevo_add_mock = responses.post(
+        "https://api.brevo.com/v3/contacts/lists/test_followup_list_name/contacts/add",
+        json={"message": "Contacts added successfully"},
+        status=201,
+    )
+
+    user = UserFactory()
+    registration = UserRegistrationRequest.objects.create(user=user)
+    activation_code = ActivationCodeFactory()
+    activation = activation_code.use(user)
+
+    registration.refresh_from_db()
+    assert registration.user_activation == activation
+
+    assert len(brevo_remove_mock.calls) == 1
+    assert brevo_remove_mock.calls[0].request.headers["api-key"] == "test_brevo_api_key"
+    assert json.loads(brevo_remove_mock.calls[0].request.body) == {"emails": [user.email]}
+
+    assert len(brevo_create_contact.calls) == 1
+    assert brevo_create_contact.calls[0].request.headers["api-key"] == "test_brevo_api_key"
+    assert json.loads(brevo_create_contact.calls[0].request.body) == {
+        "email": user.email,
+        "updateEnabled": True,
+    }
+
+    assert len(brevo_add_mock.calls) == 1
+    assert brevo_add_mock.calls[0].request.headers["api-key"] == "test_brevo_api_key"
+    assert json.loads(brevo_add_mock.calls[0].request.body) == {"emails": [user.email]}

src/backend/activation_codes/tests/test_permissions.py

@@ -0,0 +1,133 @@
+"""Tests for activation_codes permissions."""
+
+from django.test import RequestFactory
+
+import pytest
+from rest_framework.views import APIView
+
+from core.factories import UserFactory
+
+from activation_codes.factories import UserActivationFactory
+from activation_codes.permissions import IsActivatedUser
+
+
+@pytest.fixture(name="request_factory")
+def request_factory_fixture():
+    """Fixture to provide a request factory."""
+    return RequestFactory()
+
+
+@pytest.fixture(name="view")
+def view_fixture():
+    """Fixture to provide a basic view instance."""
+    return APIView()
+
+
+@pytest.mark.django_db
+def test_is_activated_user_permission_activation_not_required(request_factory, view, settings):
+    """Test that permission allows access when activation is not required."""
+    settings.ACTIVATION_REQUIRED = False
+    user = UserFactory()
+
+    request = request_factory.get("/")
+    request.user = user
+
+    permission = IsActivatedUser()
+    assert permission.has_permission(request, view) is True
+
+
+@pytest.mark.django_db
+def test_is_activated_user_permission_staff_user(request_factory, view, settings):
+    """Test that staff users always have permission."""
+    settings.ACTIVATION_REQUIRED = True
+    staff_user = UserFactory(email="staff@example.com", password="password123", is_staff=True)
+
+    request = request_factory.get("/")
+    request.user = staff_user
+
+    permission = IsActivatedUser()
+    assert permission.has_permission(request, view) is True
+
+
+@pytest.mark.django_db
+def test_is_activated_user_permission_anonymous_user(request_factory, view, settings):
+    """Test that anonymous users are allowed (handled by other permissions)."""
+    settings.ACTIVATION_REQUIRED = True
+
+    request = request_factory.get("/")
+    request.user = None
+
+    permission = IsActivatedUser()
+    assert permission.has_permission(request, view) is True
+
+
+@pytest.mark.django_db
+def test_is_activated_user_permission_activated_user(request_factory, view, settings):
+    """Test that activated users have permission."""
+    settings.ACTIVATION_REQUIRED = True
+
+    # Activate the user
+    activation = UserActivationFactory()
+
+    request = request_factory.get("/")
+    request.user = activation.user
+
+    permission = IsActivatedUser()
+    assert permission.has_permission(request, view) is True
+
+
+@pytest.mark.django_db
+def test_is_activated_user_permission_not_activated_user(request_factory, view, settings):
+    """Test that non-activated users do not have permission."""
+    settings.ACTIVATION_REQUIRED = True
+
+    user = UserFactory()
+
+    request = request_factory.get("/")
+    request.user = user
+
+    permission = IsActivatedUser()
+    assert permission.has_permission(request, view) is False
+
+
+@pytest.mark.django_db
+def test_is_activated_user_permission_custom_message():
+    """Test that permission has custom message for frontend."""
+    permission = IsActivatedUser()
+    assert permission.message == "activation-required"
+
+
+@pytest.mark.django_db
+def test_is_activated_user_object_permission(request_factory, view, settings):
+    """Test object-level permission delegates to has_permission."""
+    settings.ACTIVATION_REQUIRED = True
+
+    _user = UserFactory()
+
+    # Activate the user
+    activation = UserActivationFactory()
+
+    request = request_factory.get("/")
+    request.user = activation.user
+
+    permission = IsActivatedUser()
+    obj = object()  # Any object
+
+    # Object permission should delegate to has_permission
+    assert permission.has_object_permission(request, view, obj) is True
+
+
+@pytest.mark.django_db
+def test_is_activated_user_object_permission_not_activated(request_factory, view, settings):
+    """Test object-level permission when user is not activated."""
+    settings.ACTIVATION_REQUIRED = True
+
+    user = UserFactory()
+
+    request = request_factory.get("/")
+    request.user = user
+
+    permission = IsActivatedUser()
+    obj = object()
+
+    assert permission.has_object_permission(request, view, obj) is False

src/backend/activation_codes/tests/test_serializers.py

@@ -0,0 +1,150 @@
+"""Tests for activation_codes serializers."""
+
+import pytest
+
+from activation_codes.factories import ActivationCodeFactory, UserActivationFactory
+from activation_codes.serializers import (
+    ActivationCodeValidationSerializer,
+    ActivationStatusSerializer,
+    UserActivationSerializer,
+)
+
+
+@pytest.mark.django_db
+def test_activation_code_validation_serializer_valid_code():
+    """Test validating a valid activation code."""
+    # Create a valid activation code
+    _activation_code = ActivationCodeFactory(code="TEST1234ABCD5678")
+
+    serializer = ActivationCodeValidationSerializer(data={"code": "TEST1234ABCD5678"})
+    assert serializer.is_valid()
+    assert serializer.validated_data["code"] == "TEST1234ABCD5678"
+
+
+@pytest.mark.django_db
+def test_activation_code_validation_serializer_normalize_lowercase():
+    """Test that code is normalized to uppercase."""
+    # Create a valid activation code
+    _activation_code = ActivationCodeFactory(code="TEST1234ABCD5678")
+
+    serializer = ActivationCodeValidationSerializer(data={"code": "test1234abcd5678"})
+    assert serializer.is_valid()
+    assert serializer.validated_data["code"] == "TEST1234ABCD5678"
+
+
+@pytest.mark.django_db
+def test_activation_code_validation_serializer_normalize_with_spaces():
+    """Test that spaces are removed from code."""
+    # Create a valid activation code
+    _activation_code = ActivationCodeFactory(code="TEST1234ABCD5678")
+
+    serializer = ActivationCodeValidationSerializer(data={"code": "TEST 1234 ABCD 5678"})
+    assert serializer.is_valid()
+    assert serializer.validated_data["code"] == "TEST1234ABCD5678"
+
+
+@pytest.mark.django_db
+def test_activation_code_validation_serializer_normalize_with_dashes():
+    """Test that dashes are removed from code."""
+    # Create a valid activation code
+    _activation_code = ActivationCodeFactory(code="TEST1234ABCD5678")
+
+    serializer = ActivationCodeValidationSerializer(data={"code": "TEST-1234-ABCD-5678"})
+    assert serializer.is_valid()
+    assert serializer.validated_data["code"] == "TEST1234ABCD5678"
+
+
+@pytest.mark.django_db
+def test_activation_code_validation_serializer_normalize_mixed():
+    """Test that code with spaces, dashes and lowercase is normalized."""
+    # Create a valid activation code
+    _activation_code = ActivationCodeFactory(code="TEST1234ABCD5678")
+
+    serializer = ActivationCodeValidationSerializer(data={"code": " test-1234 abcd-5678 "})
+    assert serializer.is_valid()
+    assert serializer.validated_data["code"] == "TEST1234ABCD5678"
+
+
+@pytest.mark.django_db
+def test_activation_code_validation_serializer_missing_code():
+    """Test that code field is required."""
+    serializer = ActivationCodeValidationSerializer(data={})
+    assert not serializer.is_valid()
+    assert "code" in serializer.errors
+
+
+@pytest.mark.django_db
+def test_user_activation_serializer():
+    """Test serializing a user activation."""
+    activation = UserActivationFactory(activation_code__code="TEST1234ABCD5678")
+
+    serializer = UserActivationSerializer(activation)
+    data = serializer.data
+
+    assert "id" in data
+    assert data["code"] == "TEST1234ABCD5678"
+    assert "activated_at" in data
+    assert data["activated_at"] is not None
+
+
+@pytest.mark.django_db
+def test_user_activation_serializer_read_only_fields():
+    """Test that all fields are read-only."""
+    activation = UserActivationFactory()
+
+    serializer = UserActivationSerializer(activation)
+
+    # All fields should be in read_only_fields
+    meta = serializer.Meta
+    assert set(meta.read_only_fields) == set(meta.fields)
+
+
+@pytest.mark.django_db
+def test_activation_status_serializer_activated():
+    """Test serializing activation status for activated user."""
+    activation = UserActivationFactory(activation_code__code="TEST1234ABCD5678")
+
+    data = {"is_activated": True, "activation": activation, "requires_activation": True}
+
+    serializer = ActivationStatusSerializer(data)
+    serialized_data = serializer.data
+
+    assert serialized_data["is_activated"] is True
+    assert serialized_data["activation"] is not None
+    assert serialized_data["activation"]["code"] == "TEST1234ABCD5678"
+    assert serialized_data["requires_activation"] is True
+
+
+@pytest.mark.django_db
+def test_activation_status_serializer_not_activated():
+    """Test serializing activation status for non-activated user."""
+    data = {"is_activated": False, "activation": None, "requires_activation": True}
+
+    serializer = ActivationStatusSerializer(data)
+    serialized_data = serializer.data
+
+    assert serialized_data["is_activated"] is False
+    assert serialized_data["activation"] is None
+    assert serialized_data["requires_activation"] is True
+
+
+@pytest.mark.django_db
+def test_activation_status_serializer_activation_not_required():
+    """Test serializing activation status when activation is not required."""
+    data = {"is_activated": False, "activation": None, "requires_activation": False}
+
+    serializer = ActivationStatusSerializer(data)
+    serialized_data = serializer.data
+
+    assert serialized_data["is_activated"] is False
+    assert serialized_data["activation"] is None
+    assert serialized_data["requires_activation"] is False
+
+
+@pytest.mark.django_db
+def test_activation_status_serializer_all_fields_read_only():
+    """Test that all fields in ActivationStatusSerializer are read-only."""
+    serializer = ActivationStatusSerializer()
+
+    for field_name, field in serializer.fields.items():
+        assert field.read_only is True, f"Field {field_name} should be read-only"

src/backend/activation_codes/tests/test_viewsets.py

@@ -0,0 +1,411 @@
+"""Tests for activation_codes viewsets."""
+
+import json
+from datetime import timedelta
+from unittest.mock import patch
+
+from django.utils import timezone
+
+import pytest
+import responses
+from rest_framework import status
+
+from core.factories import UserFactory
+
+from activation_codes.factories import ActivationCodeFactory, UserActivationFactory
+from activation_codes.models import ActivationCode, UserActivation, UserRegistrationRequest
+
+
+@pytest.mark.django_db
+def test_activation_status_unauthenticated(api_client):
+    """Test that unauthenticated users cannot access status endpoint."""
+    response = api_client.get("/api/v1.0/activation/status/")
+    assert response.status_code == status.HTTP_401_UNAUTHORIZED
+
+
+@pytest.mark.django_db
+def test_activation_status_authenticated_not_activated(api_client, settings):
+    """Test activation status for authenticated but not activated user."""
+    settings.ACTIVATION_REQUIRED = True
+
+    user = UserFactory()
+    api_client.force_authenticate(user=user)
+
+    response = api_client.get("/api/v1.0/activation/status/")
+
+    assert response.status_code == status.HTTP_200_OK
+    assert response.data["is_activated"] is False
+    assert response.data["activation"] is None
+    assert response.data["requires_activation"] is True
+
+
+@pytest.mark.django_db
+def test_activation_status_authenticated_activated(api_client, settings):
+    """Test activation status for activated user."""
+    settings.ACTIVATION_REQUIRED = True
+    activation = UserActivationFactory(activation_code__code="TEST1234ABCD5678")
+    api_client.force_authenticate(user=activation.user)
+
+    response = api_client.get("/api/v1.0/activation/status/")
+
+    assert response.status_code == status.HTTP_200_OK
+    assert response.data["is_activated"] is True
+    assert response.data["activation"] is not None
+    assert response.data["activation"]["code"] == "TEST1234ABCD5678"
+    assert "activated_at" in response.data["activation"]
+    assert response.data["requires_activation"] is True
+
+
+@pytest.mark.django_db
+def test_activation_status_activation_not_required(api_client, settings):
+    """Test activation status when activation is not required."""
+    settings.ACTIVATION_REQUIRED = False
+    user = UserFactory()
+    api_client.force_authenticate(user=user)
+
+    response = api_client.get("/api/v1.0/activation/status/")
+
+    assert response.status_code == status.HTTP_200_OK
+    assert response.data["requires_activation"] is False
+
+
+@pytest.mark.django_db
+def test_validate_code_unauthenticated(api_client):
+    """Test that unauthenticated users cannot validate codes."""
+    response = api_client.post("/api/v1.0/activation/validate/", {"code": "TEST1234ABCD5678"})
+    assert response.status_code == status.HTTP_401_UNAUTHORIZED
+
+
+@pytest.mark.django_db
+def test_validate_code_success(api_client):
+    """Test successfully validating and using an activation code."""
+    user = UserFactory()
+    activation_code = ActivationCode.objects.create(code="TEST1234ABCD5678")
+    api_client.force_authenticate(user=user)
+
+    with patch("activation_codes.viewsets.logger") as mock_logger:
+        response = api_client.post("/api/v1.0/activation/validate/", {"code": "TEST1234ABCD5678"})
+
+    assert response.status_code == status.HTTP_201_CREATED
+    assert "Your account has been successfully activated" in response.data["detail"]
+    assert "activation" in response.data
+    assert response.data["activation"]["code"] == "TEST1234ABCD5678"
+
+    # Verify user is now activated
+    assert UserActivation.objects.filter(user=user).exists()
+
+    # Verify activation code was used
+    activation_code.refresh_from_db()
+    assert activation_code.current_uses == 1
+
+    # Verify logging
+    mock_logger.info.assert_called_once()
+
+
+@pytest.mark.django_db
+def test_validate_code_with_spaces_and_lowercase(api_client):
+    """Test validating code with spaces and lowercase."""
+    user = UserFactory()
+    ActivationCodeFactory(code="TEST1234ABCD5678")
+    api_client.force_authenticate(user=user)
+
+    response = api_client.post("/api/v1.0/activation/validate/", {"code": "test 1234 abcd 5678"})
+
+    assert response.status_code == status.HTTP_201_CREATED
+    assert UserActivation.objects.filter(user=user).exists()
+
+
+@pytest.mark.django_db
+def test_validate_code_already_activated(api_client):
+    """Test validating code when user is already activated."""
+    # First activation
+    activation = UserActivationFactory()
+    api_client.force_authenticate(user=activation.user)
+
+    # Try to activate again with different code
+    _another_code = ActivationCodeFactory(code="ANOTHER123456789")
+    response = api_client.post("/api/v1.0/activation/validate/", {"code": "ANOTHER123456789"})
+
+    assert response.status_code == status.HTTP_400_BAD_REQUEST
+    assert response.data == {"code": "account-already-activated"}
+
+
+@pytest.mark.django_db
+def test_validate_code_nonexistent(api_client):
+    """Test validating a non-existent code."""
+    user = UserFactory()
+    api_client.force_authenticate(user=user)
+
+    response = api_client.post("/api/v1.0/activation/validate/", {"code": "NONEXISTENT12345"})
+
+    assert response.status_code == status.HTTP_400_BAD_REQUEST
+    assert response.data == {"code": "invalid-code"}
+
+
+@pytest.mark.django_db
+def test_validate_code_invalid_serializer(api_client):
+    """Test validating with invalid data."""
+    user = UserFactory()
+    api_client.force_authenticate(user=user)
+
+    response = api_client.post(
+        "/api/v1.0/activation/validate/",
+        {},  # Missing code
+    )
+
+    assert response.status_code == status.HTTP_400_BAD_REQUEST
+    assert "code" in response.data
+
+
+@pytest.mark.django_db
+def test_validate_code_inactive(api_client):
+    """Test validating an inactive code."""
+    user = UserFactory()
+    ActivationCodeFactory(code="INACTIVE12345678", is_active=False)
+    api_client.force_authenticate(user=user)
+
+    response = api_client.post("/api/v1.0/activation/validate/", {"code": "INACTIVE12345678"})
+
+    assert response.status_code == status.HTTP_400_BAD_REQUEST
+    assert response.data == {"code": "invalid-code"}
+
+
+@pytest.mark.django_db
+def test_validate_code_expired(api_client):
+    """Test validating an expired code."""
+    user = UserFactory()
+    ActivationCodeFactory(code="EXPIRED123456789", expires_at=timezone.now() - timedelta(days=1))
+    api_client.force_authenticate(user=user)
+
+    response = api_client.post("/api/v1.0/activation/validate/", {"code": "EXPIRED123456789"})
+
+    assert response.status_code == status.HTTP_400_BAD_REQUEST
+
+
+@pytest.mark.django_db
+def test_validate_code_max_uses_reached(api_client):
+    """Test validating a code that has reached max uses."""
+    user = UserFactory()
+    ActivationCodeFactory(code="MAXUSED123456789", max_uses=1, current_uses=1)
+    api_client.force_authenticate(user=user)
+
+    response = api_client.post("/api/v1.0/activation/validate/", {"code": "MAXUSED123456789"})
+
+    assert response.status_code == status.HTTP_400_BAD_REQUEST
+
+
+@pytest.mark.django_db
+def test_validate_code_multi_use(api_client):
+    """Test using a multi-use code with multiple users."""
+    multi_use_activation_code = ActivationCodeFactory(
+        code="MULTIUSE12345678",
+        max_uses=3,
+    )
+    users = []
+    for i in range(3):
+        user = UserFactory(email=f"user{i}@example.com")
+        users.append(user)
+
+    for i, user in enumerate(users):
+        api_client.force_authenticate(user=user)
+        response = api_client.post("/api/v1.0/activation/validate/", {"code": "MULTIUSE12345678"})
+
+        assert response.status_code == status.HTTP_201_CREATED
+        multi_use_activation_code.refresh_from_db()
+        assert multi_use_activation_code.current_uses == i + 1
+
+
+@pytest.mark.django_db
+def test_validate_code_unlimited_use(api_client):
+    """Test using an unlimited code with multiple users."""
+    unlimited_activation_code = ActivationCodeFactory(
+        code="UNLIMITED123CODE",
+        max_uses=0,  # Unlimited uses
+    )
+    for i in range(10):
+        user = UserFactory(email=f"user{i}@example.com")
+        api_client.force_authenticate(user=user)
+        response = api_client.post("/api/v1.0/activation/validate/", {"code": "UNLIMITED123CODE"})
+
+        assert response.status_code == status.HTTP_201_CREATED
+
+    # Code should still be valid
+    unlimited_activation_code.refresh_from_db()
+    assert unlimited_activation_code.is_valid() is True
+
+
+@pytest.mark.django_db
+def test_validate_code_logging_on_validation_error(api_client):
+    """Test that validation errors are logged."""
+    user = UserFactory()
+    api_client.force_authenticate(user=user)
+
+    # Create a code that will cause validation error
+    code = ActivationCodeFactory(
+        code="WILLEXPIRE123456", expires_at=timezone.now() + timedelta(days=1)
+    )
+
+    # Make it expire
+    code.expires_at = timezone.now() - timedelta(seconds=1)
+    code.save()
+
+    with patch("activation_codes.viewsets.logger"):
+        response = api_client.post("/api/v1.0/activation/validate/", {"code": "WILLEXPIRE123456"})
+
+    assert response.status_code == status.HTTP_400_BAD_REQUEST
+
+    # Note: In this case the serializer will catch it first
+    # so the warning might not be called, but this tests the flow
+
+
+@pytest.mark.django_db
+def test_unauthenticated_register_email(api_client):
+    """Test that unauthenticated users cannot register email."""
+    response = api_client.post(
+        "/api/v1.0/activation/register/",
+        {
+            "email": "test@example.com",
+        },
+    )
+    assert response.status_code == status.HTTP_401_UNAUTHORIZED
+
+
+@pytest.mark.django_db
+def test_register_email_success(api_client):
+    """Test successfully registering an email."""
+    user = UserFactory()
+    api_client.force_authenticate(user=user)
+
+    response = api_client.post(
+        "/api/v1.0/activation/register/",
+        {},
+    )
+    assert response.status_code == status.HTTP_201_CREATED
+    assert response.data["code"] == "registration-successful"
+
+    registration = UserRegistrationRequest.objects.get(user=user)
+    assert registration.user == user
+
+
+@pytest.mark.django_db
+def test_register_already_created(api_client):
+    """Test successfully registering an email."""
+    user = UserFactory()
+    _registration = UserRegistrationRequest.objects.create(
+        user=user,
+    )
+
+    api_client.force_authenticate(user=user)
+
+    response = api_client.post(
+        "/api/v1.0/activation/register/",
+    )
+    assert response.status_code == status.HTTP_200_OK
+    assert response.data == {"code": "registration-successful"}
+
+    assert UserRegistrationRequest.objects.filter(user=user).count() == 1
+
+
+@pytest.mark.django_db
+def test_validate_code_registered_user(api_client):
+    """Test validating a code for a user with a pre-existing registration."""
+    user = UserFactory()
+    _registration = UserRegistrationRequest.objects.create(
+        user=user,
+    )
+    activation_code = ActivationCodeFactory(code="TEST1234ABCD5678")
+
+    api_client.force_authenticate(user=user)
+
+    response = api_client.post("/api/v1.0/activation/validate/", {"code": "TEST1234ABCD5678"})
+
+    assert response.status_code == status.HTTP_201_CREATED
+
+    _registration.refresh_from_db()
+    assert _registration.user_activation.activation_code == activation_code
+
+
+@responses.activate
+@pytest.mark.django_db
+def test_register_email_success_brevo(api_client, settings):
+    """Test successfully registering an email and notify Brevo."""
+    settings.BREVO_API_KEY = "test_brevo_api_key"
+    settings.BREVO_WAITING_LIST_ID = "test_waiting_list_id"
+
+    brevo_create_contact = responses.post(
+        "https://api.brevo.com/v3/contacts",
+        status=200,
+    )
+
+    brevo_mock = responses.post(
+        "https://api.brevo.com/v3/contacts/lists/test_waiting_list_id/contacts/add",
+        json={"message": "Contacts added successfully"},
+        status=201,
+    )
+
+    user = UserFactory()
+    api_client.force_authenticate(user=user)
+
+    response = api_client.post(
+        "/api/v1.0/activation/register/",
+        {},
+    )
+    assert response.status_code == status.HTTP_201_CREATED
+    assert response.data["code"] == "registration-successful"
+
+    registration = UserRegistrationRequest.objects.get(user=user)
+    assert registration.user == user
+
+    assert len(brevo_create_contact.calls) == 1
+    assert brevo_create_contact.calls[0].request.headers["api-key"] == "test_brevo_api_key"
+    assert json.loads(brevo_create_contact.calls[0].request.body) == {
+        "email": user.email,
+        "updateEnabled": True,
+    }
+
+    assert len(brevo_mock.calls) == 1
+    assert brevo_mock.calls[0].request.headers["api-key"] == "test_brevo_api_key"
+    assert json.loads(brevo_mock.calls[0].request.body) == {"emails": [user.email]}
+
+    # Register again to test idempotency
+    response = api_client.post(
+        "/api/v1.0/activation/register/",
+        {},
+    )
+    assert response.status_code == status.HTTP_200_OK
+    assert response.data["code"] == "registration-successful"
+
+    assert len(brevo_mock.calls) == 1  # No new call made
+
+
+@responses.activate
+@pytest.mark.django_db
+def test_register_email_success_brevo_fails(api_client, settings):
+    """Test successfully registering an email, even if Brevo fails."""
+    settings.BREVO_API_KEY = "test_brevo_api_key"
+    settings.BREVO_WAITING_LIST_ID = "test_waiting_list_id"
+
+    _brevo_create_contact = responses.post(
+        "https://api.brevo.com/v3/contacts",
+        status=200,
+    )
+
+    brevo_mock = responses.post(
+        "https://api.brevo.com/v3/contacts/lists/test_waiting_list_id/contacts/add",
+        status=400,
+    )
+
+    user = UserFactory()
+    api_client.force_authenticate(user=user)
+
+    response = api_client.post(
+        "/api/v1.0/activation/register/",
+        {},
+    )
+    assert response.status_code == status.HTTP_201_CREATED
+    assert response.data["code"] == "registration-successful"
+
+    registration = UserRegistrationRequest.objects.get(user=user)
+    assert registration.user == user
+
+    assert len(brevo_mock.calls) == 1

src/backend/activation_codes/viewsets.py

@@ -0,0 +1,153 @@
+"""API ViewSets for activation codes."""
+
+import logging
+
+from django.conf import settings
+from django.core.exceptions import ValidationError
+from django.utils.translation import gettext_lazy as _
+
+from rest_framework import status, viewsets
+from rest_framework.decorators import action
+from rest_framework.response import Response
+
+from core.brevo import add_user_to_brevo_list
+from core.permissions import IsAuthenticated
+
+from . import models, serializers
+from .exceptions import InvalidCodeError, UserAlreadyActivatedError
+
+logger = logging.getLogger(__name__)
+
+
+class ActivationViewSet(viewsets.GenericViewSet):
+    """
+    ViewSet for handling user activation with codes.
+
+    Endpoints:
+    - GET /activation/status/ - Check if current user is activated
+    - POST /activation/validate/ - Validate and use an activation code
+    - POST /activation/register/ - Register an email to be notified later
+    """
+
+    permission_classes = [IsAuthenticated]
+    serializer_class = serializers.ActivationCodeValidationSerializer
+
+    @action(detail=False, methods=["get"], url_path="status")
+    def status(self, request):
+        """
+        Get the activation status of the current user.
+
+        Returns:
+            - is_activated: Whether the user has activated their account
+            - activation: Details of the activation (if exists)
+            - requires_activation: Whether activation is required by the system
+        """
+        requires_activation = getattr(settings, "ACTIVATION_REQUIRED", False)
+
+        try:
+            activation = models.UserActivation.objects.select_related("activation_code").get(
+                user=request.user
+            )
+            is_activated = True
+        except models.UserActivation.DoesNotExist:
+            activation = None
+            is_activated = False
+
+        response_data = {
+            "is_activated": is_activated,
+            "activation": activation,
+            "requires_activation": requires_activation,
+        }
+
+        return Response(
+            serializers.ActivationStatusSerializer(response_data).data, status=status.HTTP_200_OK
+        )
+
+    @action(detail=False, methods=["post"], url_path="validate")
+    def validate_code(self, request):
+        """
+        Validate an activation code and activate the user's account.
+
+        Request body:
+            - code: The activation code to validate
+
+        Returns:
+            - Success: Activation details
+            - Error: Validation error message
+        """
+        serializer = self.get_serializer(data=request.data)
+        serializer.is_valid(raise_exception=True)
+
+        code_value = serializer.validated_data["code"]
+
+        # Get the activation code
+        try:
+            activation_code = models.ActivationCode.objects.get(code=code_value)
+        except models.ActivationCode.DoesNotExist:
+            logger.info("Activation code %s does not exist", code_value)
+            return Response({"code": "invalid-code"}, status=status.HTTP_400_BAD_REQUEST)
+
+        # Use the code
+        try:
+            activation = activation_code.use(request.user)
+        except InvalidCodeError as exc:
+            logger.warning(exc)
+            return Response({"code": "invalid-code"}, status=status.HTTP_400_BAD_REQUEST)
+        except UserAlreadyActivatedError as exc:
+            logger.info(exc)
+            return Response(
+                {"code": "account-already-activated"}, status=status.HTTP_400_BAD_REQUEST
+            )
+
+        logger.info("User %s activated account with code %s", request.user.id, activation_code.code)
+
+        return Response(
+            {
+                "code": "activation-successful",
+                "detail": _("Your account has been successfully activated"),
+                "activation": serializers.UserActivationSerializer(activation).data,
+            },
+            status=status.HTTP_201_CREATED,
+        )
+
+    @action(detail=False, methods=["post"], url_path="register")
+    def register_email(self, request):
+        """
+        Register an email to be notified when activation codes are available.
+
+        Request body:
+            - email: The email address to register
+
+        Returns:
+            - Success: Confirmation message
+            - Error: Validation error message
+        """
+        serializer = serializers.UserRegistrationRequestSerializer(
+            data={},
+            context={"request": request},
+        )
+        serializer.is_valid(raise_exception=True)
+
+        # Create the registration
+        try:
+            serializer.save()
+        except ValidationError:
+            # user is already registered, it's OK
+            return Response(
+                {"code": "registration-successful"},
+                status=status.HTTP_200_OK,
+            )
+
+        add_user_to_brevo_list(
+            [serializer.validated_data["user"].email], settings.BREVO_WAITING_LIST_ID
+        )
+
+        logger.info(
+            "Registered email %s for activation notifications",
+            serializer.validated_data["user"].email,
+        )
+
+        return Response(
+            {"code": "registration-successful"},
+            status=status.HTTP_201_CREATED,
+        )

src/backend/chat/admin.py

@@ -9,8 +9,28 @@ from . import models
 class ChatConversationAdmin(admin.ModelAdmin):
     """Admin class for the ChatConversation model"""
 
+    autocomplete_fields = ("owner", "project")
+    list_select_related = ("project",)
+
     list_display = (
         "id",
+        "title",
+        "project",
+        "created_at",
+        "updated_at",
+    )
+
+
+@admin.register(models.ChatProject)
+class ChatProjectAdmin(admin.ModelAdmin):
+    """Admin class for the ChatProject model"""
+
+    search_fields = ("title",)
+    list_display = (
+        "id",
+        "title",
+        "icon",
+        "color",
         "created_at",
         "updated_at",
     )

src/backend/chat/agent_rag/albert_api_constants.py

@@ -0,0 +1,150 @@
+"""Constants and schemas for the Albert RAG agent from Albert API codebase."""
+
+from enum import Enum
+from typing import Annotated, Any, Dict, List, Literal, Optional, Self
+
+from pydantic import BaseModel, Field, StringConstraints, model_validator
+
+
+# - app/schemas/chunks.py
+class Chunk(BaseModel):
+    """Model representing a chunk of text with metadata."""
+
+    object: Literal["chunk"] = "chunk"
+    id: int
+    metadata: Dict[str, Any]
+    content: str
+
+
+class Chunks(BaseModel):
+    """Model representing a list of chunks."""
+
+    object: Literal["list"] = "list"
+    data: List[Chunk]
+
+
+# - app/schemas/usage.py
+class CarbonFootprintUsageKWh(BaseModel):
+    """Model representing the carbon footprint usage in kWh (kilowatt-hours)."""
+
+    min: Optional[float] = Field(default=None, description="Minimum carbon footprint in kWh.")
+    max: Optional[float] = Field(default=None, description="Maximum carbon footprint in kWh.")
+
+
+class CarbonFootprintUsageKgCO2eq(BaseModel):
+    """Model representing the carbon footprint usage in kgCO2eq (kilograms of CO2 equivalent)."""
+
+    min: Optional[float] = Field(
+        default=None, description="Minimum carbon footprint in kgCO2eq (global warming potential)."
+    )
+    max: Optional[float] = Field(
+        default=None, description="Maximum carbon footprint in kgCO2eq (global warming potential)."
+    )
+
+
+class CarbonFootprintUsage(BaseModel):
+    """Model representing the carbon footprint usage in kWh and kgCO2eq."""
+
+    kWh: CarbonFootprintUsageKWh = Field(default_factory=CarbonFootprintUsageKWh)
+    kgCO2eq: CarbonFootprintUsageKgCO2eq = Field(default_factory=CarbonFootprintUsageKgCO2eq)
+
+
+class BaseUsage(BaseModel):
+    """Base model for usage statistics in the Albert API."""
+
+    prompt_tokens: int = Field(
+        default=0, description="Number of prompt tokens (e.g. input tokens)."
+    )
+    completion_tokens: int = Field(
+        default=0, description="Number of completion tokens (e.g. output tokens)."
+    )
+    total_tokens: int = Field(
+        default=0, description="Total number of tokens (e.g. input and output tokens)."
+    )
+    cost: float = Field(default=0.0, description="Total cost of the request.")
+    carbon: CarbonFootprintUsage = Field(default_factory=CarbonFootprintUsage)
+
+
+# - app/schemas/usage.py
+class Detail(BaseModel):
+    """Model representing a detail in the usage statistics."""
+
+    id: str
+    model: str
+    usage: BaseUsage = Field(default_factory=BaseUsage)
+
+
+class Usage(BaseUsage):
+    """Model representing the usage statistics for the Albert API."""
+
+    details: List[Detail] = []
+
+
+class SearchMethod(str, Enum):
+    """
+    Enum representing the search methods available (will be displayed in this order in playground).
+    """
+
+    MULTIAGENT = "multiagent"
+    HYBRID = "hybrid"
+    SEMANTIC = "semantic"
+    LEXICAL = "lexical"
+
+
+class SearchArgs(BaseModel):
+    """Model representing the arguments for a search request in the Albert API."""
+
+    collections: List[Any] = Field(default=[], description="List of collections ID")
+    rff_k: int = Field(default=20, description="k constant in RFF algorithm")
+    k: int = Field(gt=0, default=4, description="Number of results to return")
+    method: SearchMethod = Field(default=SearchMethod.SEMANTIC)
+    score_threshold: Optional[float] = Field(
+        default=0.0,
+        ge=0.0,
+        le=1.0,
+        description=(
+            "Score of cosine similarity threshold for filtering results, "
+            "only available for semantic search method."
+        ),
+    )
+    web_search: bool = Field(
+        default=False, description="Whether add internet search to the results."
+    )
+    web_search_k: int = Field(default=5, description="Number of results to return for web search.")
+
+    @model_validator(mode="after")
+    def score_threshold_filter(self) -> Self:
+        """Validate the score threshold based on the search method."""
+        if self.score_threshold and self.method not in (
+            SearchMethod.SEMANTIC,
+            SearchMethod.MULTIAGENT,
+        ):
+            raise ValueError(
+                "Score threshold is only available for semantic and multiagent search methods."
+            )
+        return self
+
+
+class SearchRequest(SearchArgs):
+    """Model representing a search request in the Albert API."""
+
+    prompt: Annotated[
+        str,
+        StringConstraints(strip_whitespace=True, min_length=1),
+    ] = Field(description="Prompt related to the search")
+
+
+class Search(BaseModel):
+    """Model representing a search result in the Albert API."""
+
+    method: SearchMethod
+    score: float
+    chunk: Chunk
+
+
+class Searches(BaseModel):
+    """Model representing a list of search results in the Albert API."""
+
+    object: Literal["list"] = "list"
+    data: List[Search]
+    usage: Usage = Field(default_factory=Usage, description="Usage information for the request.")

src/backend/chat/agent_rag/constants.py

@@ -0,0 +1,41 @@
+"""Constants for RAG (Retrieval-Augmented Generation) results."""
+
+from typing import List
+
+from pydantic import BaseModel, Field
+
+
+class RAGWebUsage(BaseModel):
+    """
+    Model representing the usage statistics for web results in RAG (Retrieval-Augmented Generation).
+    """
+
+    prompt_tokens: int = Field(default=0, description="Number of prompt tokens used.")
+    completion_tokens: int = Field(default=0, description="Number of completion tokens generated.")
+
+
+class RAGWebResult(BaseModel):
+    """Model representing a single web result in RAG (Retrieval-Augmented Generation)."""
+
+    url: str = Field(..., description="URL of the web result.")
+    content: str = Field(..., description="Content of the web result chunk.")
+    score: float = Field(
+        ..., description="Relevance score of the web result, typically between 0 and 1."
+    )
+
+
+class RAGWebResults(BaseModel):
+    """Model representing a list of web results in RAG (Retrieval-Augmented Generation)."""
+
+    data: List[RAGWebResult]
+    usage: RAGWebUsage = Field(..., description="RAG usage statistics.")
+
+    def to_prompt(self) -> str:
+        """Convert the web results to a prompt string."""
+        _format = " - From: {url}:\n   content: {content}\n\n"
+        return (
+            "\n\n".join(
+                _format.format(url=result.url, content=result.content) for result in self.data
+            )
+            + "\n\n"
+        )

src/backend/chat/agent_rag/document_converter/markitdown.py

@@ -0,0 +1,43 @@
+"""Document Converter using MarkItDown"""
+
+import os.path
+from io import BytesIO
+
+from markitdown import MarkItDown
+
+
+class DocumentConverter:
+    """Simple document converter that uses MarkItDown to convert documents to Markdown format."""
+
+    def __init__(self):
+        """Initialize the DocumentConverter with MarkItDown."""
+        self.converter = MarkItDown()
+
+    def convert_raw(  # pylint: disable=unused-argument
+        self,
+        *,
+        name: str,
+        content_type: str,
+        content: bytes,
+    ) -> str:
+        """
+        Convert a document to Markdown format.
+        The name, content_type, and content parameters comes from the user input
+        (vercel SDK Attachment, or BinaryContent).
+
+        Args:
+            name (str): The name of the document.
+            content_type (str): The MIME type of the document (e.g., "application/pdf").
+            content (bytes): The content of the document as bytes.
+        """
+        return self._convert(BytesIO(content), file_extension=os.path.splitext(name)[1])
+
+    def _convert(self, document: BytesIO, file_extension: str) -> str:
+        """
+        Convert the given document using the underlying DocumentConverter.
+        """
+        conversion = self.converter.convert_stream(
+            document, file_extension=file_extension or ".txt"
+        )
+        document_markdown = conversion.text_content
+        return document_markdown

src/backend/chat/agent_rag/document_converter/parser.py

@@ -0,0 +1,280 @@
+"""Document parsers for RAG backends."""
+
+import base64
+import logging
+import time
+from io import BytesIO
+from urllib.parse import urljoin
+
+from django.conf import settings
+
+import requests
+from pypdf import PdfReader, PdfWriter
+
+from chat.agent_rag.document_converter.markitdown import DocumentConverter
+
+logger = logging.getLogger(__name__)
+
+
+class BaseParser:
+    """Base class for document parsers."""
+
+    def parse_document(self, name: str, content_type: str, content: bytes) -> str:
+        """
+        Parse the document and prepare it for the search operation.
+        This method should handle the logic to convert the document
+        into a format suitable for storage.
+
+        Args:
+            name (str): The name of the document.
+            content_type (str): The MIME type of the document (e.g., "application/pdf").
+            content (bytes): The content of the document as a bytes stream.
+
+        Returns:
+            str: The document content in Markdown format.
+        """
+        raise NotImplementedError("Must be implemented in subclass.")
+
+
+class AlbertParser(BaseParser):
+    """Document parser using Albert API for PDFs and DocumentConverter for other formats."""
+
+    endpoint = urljoin(settings.ALBERT_API_URL, "/v1/parse-beta")
+
+    def parse_pdf_document(self, name: str, content_type: str, content: bytes) -> str:
+        """Parse PDF document using Albert API."""
+        response = requests.post(
+            self.endpoint,
+            headers={
+                "Authorization": f"Bearer {settings.ALBERT_API_KEY}",
+            },
+            files={
+                "file": (name, content, content_type),
+                "output_format": (None, "markdown"),
+            },
+            timeout=settings.ALBERT_API_PARSE_TIMEOUT,
+        )
+        response.raise_for_status()
+
+        return "\n\n".join(
+            document_page["content"] for document_page in response.json().get("data", [])
+        )
+
+    def parse_document(self, name: str, content_type: str, content: bytes) -> str:
+        """Parse document based on content type."""
+        if content_type == "application/pdf":
+            return self.parse_pdf_document(name=name, content_type=content_type, content=content)
+        return DocumentConverter().convert_raw(
+            name=name, content_type=content_type, content=content
+        )
+
+
+METHOD_TEXT_EXTRACTION = "text_extraction"
+METHOD_OCR = "ocr"
+
+
+def analyze_pdf(pdf_data: bytes) -> dict:
+    """
+    Analyze a PDF to determine if it needs OCR or can use direct text extraction.
+    """
+    reader = PdfReader(BytesIO(pdf_data))
+    total_pages = len(reader.pages)
+    if total_pages == 0:
+        logger.info("No page found in pdf")
+        return {
+            "total_pages": 0,
+            "pages_with_text": 0,
+            "avg_chars_per_page": 0,
+            "text_coverage": 0,
+            "recommended_method": METHOD_TEXT_EXTRACTION,
+        }
+
+    total_chars = 0
+    pages_with_text = 0
+    for page in reader.pages:
+        text = (page.extract_text() or "").strip()
+        char_count = len(text)
+        total_chars += char_count
+
+        if char_count > 50:
+            pages_with_text += 1
+
+    avg_chars = total_chars / total_pages
+    text_coverage = pages_with_text / total_pages
+
+    # Decision logic
+    if (
+        avg_chars > settings.MIN_AVG_CHARS_FOR_TEXT_EXTRACTION
+        and text_coverage > settings.MIN_TEXT_COVERAGE_FOR_TEXT_EXTRACTION
+    ):
+        method = METHOD_TEXT_EXTRACTION
+
+    else:
+        method = METHOD_OCR
+
+    return {
+        "total_pages": total_pages,
+        "pages_with_text": pages_with_text,
+        "avg_chars_per_page": round(avg_chars),
+        "text_coverage": round(text_coverage, 2),
+        "recommended_method": method,
+    }
+
+
+class AdaptiveParserMixin:
+    """
+    Mixin that adds adaptive PDF parsing behavior.
+
+    Analyzes PDF content to choose between direct text extraction (fast) and OCR
+    (for scanned/image PDFs). Subclasses must implement `parse_pdf_document_with_ocr`.
+    """
+
+    def parse_pdf_document(self, name: str, content_type: str, content: bytes) -> str:
+        """Analyze PDF and route to text extraction or OCR based on content."""
+        analysis = analyze_pdf(content)
+
+        logger.info(
+            "Pdf analysis - pages: %s, pages with text: %s, text_coverage: %s, "
+            "recommended method: %s",
+            analysis["total_pages"],
+            analysis["pages_with_text"],
+            analysis["text_coverage"],
+            analysis["recommended_method"],
+        )
+
+        method = analysis["recommended_method"]
+        if method == METHOD_TEXT_EXTRACTION:
+            return self.extract_text_from_pdf(name=name, content_type=content_type, content=content)
+        return self.parse_pdf_document_with_ocr(name=name, content=content)
+
+    def extract_text_from_pdf(self, name: str, content_type: str, content: bytes) -> str:
+        """Extract text directly from PDF without OCR (for text-based PDFs)."""
+        logger.info("Parsing pdf with text extraction")
+        return DocumentConverter().convert_raw(
+            name=name, content_type=content_type, content=content
+        )
+
+    def parse_pdf_document_with_ocr(self, name: str, content: bytes) -> str:
+        """Process PDF through OCR. Must be implemented by subclass."""
+        raise NotImplementedError("Subclass must implement parse_pdf_document_with_ocr")
+
+
+class AdaptivePdfParser(AdaptiveParserMixin, BaseParser):
+    """
+    PDF parser with adaptive text extraction / OCR routing.
+
+    Uses Mistral OCR API for scanned/image PDFs, processed in batches with retry logic.
+    """
+
+    def __init__(self):
+        super().__init__()
+
+        self.endpoint = urljoin(
+            settings.LLM_CONFIGURATIONS[settings.OCR_HRID].provider.base_url, "/v1/ocr"
+        )
+        self.max_retries = settings.OCR_MAX_RETRIES
+        self.retry_delay = settings.OCR_RETRY_DELAY
+        api_key = settings.LLM_CONFIGURATIONS[settings.OCR_HRID].provider.api_key
+
+        self.headers = {
+            "Authorization": f"Bearer {api_key}",
+            "Content-Type": "application/json",
+        }
+
+    def extract_page_batch(self, reader: PdfReader, start_index: int, end_index: int) -> bytes:
+        """Extract a range of pages from PDF as a new PDF bytes object."""
+        writer = PdfWriter()
+        for i in range(start_index, end_index):
+            writer.add_page(reader.pages[i])
+        output = BytesIO()
+        writer.write(output)
+        return output.getvalue()
+
+    def ocr_page_batch(
+        self,
+        name: str,
+        page_content: bytes,
+        start_index: int,
+        end_index: int,
+    ) -> list[str]:
+        """Send page batch to Mistral OCR API with static delay retry."""
+        file_data = base64.standard_b64encode(page_content).decode("utf-8")
+        payload = {
+            "document": {
+                "type": "document_url",
+                "document_name": f"{name}_pages_{start_index + 1}_to_{end_index}",
+                "document_url": f"data:application/pdf;base64,{file_data}",
+            },
+            "model": settings.OCR_MODEL,
+        }
+
+        last_exception = None
+        for attempt in range(self.max_retries):
+            try:
+                response = requests.post(
+                    self.endpoint,
+                    headers=self.headers,
+                    json=payload,
+                    timeout=settings.OCR_TIMEOUT,
+                )
+                response.raise_for_status()
+
+                pages = response.json().get("pages", [])
+                return [page.get("markdown", "") for page in pages]
+
+            except (requests.Timeout, requests.RequestException) as e:
+                last_exception = e
+                if attempt < self.max_retries - 1:
+                    logger.warning(
+                        "OCR attempt %d/%d failed for pages %d-%d: %s. Retrying in %.1fs...",
+                        attempt + 1,
+                        self.max_retries,
+                        start_index + 1,
+                        end_index,
+                        str(e),
+                        self.retry_delay,
+                    )
+                    time.sleep(self.retry_delay)
+
+        logger.error(
+            "OCR failed for pages %d-%d after %d attempts: %s",
+            start_index + 1,
+            end_index,
+            self.max_retries,
+            str(last_exception),
+        )
+        raise last_exception
+
+    def parse_pdf_document_with_ocr(self, name: str, content: bytes) -> str:
+        """Process PDF through OCR in batches, returning concatenated markdown."""
+        reader = PdfReader(BytesIO(content))
+        total_pages = len(reader.pages)
+        batch_size = settings.OCR_BATCH_PAGES
+
+        logger.info("Parsing pdf with OCR (%d pages, batch size %d)", total_pages, batch_size)
+
+        results = []
+        for start_index in range(0, total_pages, batch_size):
+            end_index = min(start_index + batch_size, total_pages)
+            batch_content = self.extract_page_batch(reader, start_index, end_index)
+            try:
+                batch_results = self.ocr_page_batch(name, batch_content, start_index, end_index)
+                results.extend(batch_results)
+                logger.debug(
+                    "Completed OCR for pages %d-%d/%d", start_index + 1, end_index, total_pages
+                )
+            except Exception as e:  # pylint: disable=broad-except #noqa: BLE001
+                logger.error("Failed to OCR pages %d-%d: %s", start_index + 1, end_index, str(e))
+                # Preserve page count with empty placeholders to maintain correct ordering
+                results.extend([""] * (end_index - start_index))
+
+        return "\n\n".join(results)
+
+    def parse_document(self, name: str, content_type: str, content: bytes) -> str:
+        """Route to PDF parser or DocumentConverter based on content type."""
+        if content_type == "application/pdf":
+            return self.parse_pdf_document(name=name, content_type=content_type, content=content)
+
+        return DocumentConverter().convert_raw(
+            name=name, content_type=content_type, content=content
+        )

src/backend/chat/agent_rag/document_rag_backends/albert_rag_backend.py

@@ -0,0 +1,258 @@
+"""Implementation of the Albert API for RAG document search."""
+
+import json
+import logging
+from io import BytesIO
+from typing import List, Optional
+from urllib.parse import urljoin
+
+from django.conf import settings
+from django.utils.module_loading import import_string
+
+import httpx
+import requests
+
+from chat.agent_rag.albert_api_constants import Searches
+from chat.agent_rag.constants import RAGWebResult, RAGWebResults, RAGWebUsage
+from chat.agent_rag.document_rag_backends.base_rag_backend import BaseRagBackend
+
+logger = logging.getLogger(__name__)
+
+
+class AlbertRagBackend(BaseRagBackend):  # pylint: disable=too-many-instance-attributes
+    """
+    This class is a placeholder for the Albert API implementation.
+    It is designed to be used with the RAG (Retrieval-Augmented Generation) document search system.
+
+    It provides methods to:
+    - Create a collection for the search operation.
+    - Store parsed documents in the Albert collection.
+    - Perform a search operation using the Albert API.
+    """
+
+    def __init__(
+        self,
+        collection_id: Optional[str] = None,
+        read_only_collection_id: Optional[List[str]] = None,
+    ):
+        # Initialize any necessary parameters or configurations here
+        super().__init__(collection_id, read_only_collection_id)
+        self._base_url = settings.ALBERT_API_URL
+        self._headers = {
+            "Authorization": f"Bearer {settings.ALBERT_API_KEY}",
+        }
+        self._collections_endpoint = urljoin(self._base_url, "/v1/collections")
+        self._documents_endpoint = urljoin(self._base_url, "/v1/documents")
+        self._search_endpoint = urljoin(self._base_url, "/v1/search")
+        self._default_collection_description = "Temporary collection for RAG document search"
+        parser_class = import_string(settings.RAG_DOCUMENT_PARSER)
+        self.parser = parser_class()
+
+    @staticmethod
+    def cast_collection_id(collection_id):
+        """Albert API expects int Ids."""
+        return int(collection_id)
+
+    def create_collection(self, name: str, description: Optional[str] = None) -> str:
+        """
+        Create a temporary collection for the search operation.
+        This method should handle the logic to create or retrieve an existing collection.
+        """
+        response = requests.post(
+            self._collections_endpoint,
+            headers=self._headers,
+            json={
+                "name": name,
+                "description": description or self._default_collection_description,
+                "visibility": "private",
+            },
+            timeout=settings.ALBERT_API_TIMEOUT,
+        )
+        response.raise_for_status()
+        self.collection_id = str(response.json()["id"])
+        return self.collection_id
+
+    async def acreate_collection(self, name: str, description: Optional[str] = None) -> str:
+        """
+        Create a temporary collection for the search operation.
+        This method should handle the logic to create or retrieve an existing collection.
+        """
+        async with httpx.AsyncClient(timeout=settings.ALBERT_API_TIMEOUT) as client:
+            response = await client.post(
+                self._collections_endpoint,
+                headers=self._headers,
+                json={
+                    "name": name,
+                    "description": description or self._default_collection_description,
+                    "visibility": "private",
+                },
+                timeout=settings.ALBERT_API_TIMEOUT,
+            )
+            response.raise_for_status()
+
+        self.collection_id = str(response.json()["id"])
+        return self.collection_id
+
+    def delete_collection(self, **kwargs) -> None:
+        """
+        Delete the current collection
+        """
+        response = requests.delete(
+            urljoin(f"{self._collections_endpoint}/", self.collection_id),
+            headers=self._headers,
+            timeout=settings.ALBERT_API_TIMEOUT,
+        )
+        response.raise_for_status()
+
+    async def adelete_collection(self, **kwargs) -> None:
+        """
+        Asynchronously delete the current collection
+        """
+        async with httpx.AsyncClient(timeout=settings.ALBERT_API_TIMEOUT) as client:
+            response = await client.delete(
+                urljoin(f"{self._collections_endpoint}/", self.collection_id),
+                headers=self._headers,
+                timeout=settings.ALBERT_API_TIMEOUT,
+            )
+            response.raise_for_status()
+
+    def store_document(self, name: str, content: str, **kwargs) -> None:
+        """
+        Store the document content in the Albert collection.
+        This method should handle the logic to send the document content to the Albert API.
+
+        Args:
+            name (str): The name of the document.
+            content (str): The content of the document in Markdown format.
+            **kwargs: Additional arguments.
+        """
+        response = requests.post(
+            urljoin(self._base_url, self._documents_endpoint),
+            headers=self._headers,
+            files={
+                "file": (f"{name}.md", BytesIO(content.encode("utf-8")), "text/markdown"),
+                "collection": (None, int(self.collection_id)),
+                "metadata": (None, json.dumps({"document_name": name})),  # undocumented API
+            },
+            timeout=settings.ALBERT_API_TIMEOUT,
+        )
+        logger.debug(response.json())
+        response.raise_for_status()
+
+    async def astore_document(self, name: str, content: str, **kwargs) -> None:
+        """
+        Store the document content in the Albert collection.
+        This method should handle the logic to send the document content to the Albert API.
+
+        Args:
+            name (str): The name of the document.
+            content (str): The content of the document in Markdown format.
+            **kwargs: Additional arguments.
+        """
+        async with httpx.AsyncClient(timeout=settings.ALBERT_API_TIMEOUT) as client:
+            response = await client.post(
+                urljoin(self._base_url, self._documents_endpoint),
+                headers=self._headers,
+                files={
+                    "file": (f"{name}.md", BytesIO(content.encode("utf-8")), "text/markdown"),
+                },
+                data={
+                    "collection": int(self.collection_id),
+                    "metadata": json.dumps({"document_name": name}),  # undocumented API
+                },
+                timeout=settings.ALBERT_API_TIMEOUT,
+            )
+            logger.debug(response.json())
+            response.raise_for_status()
+
+    def search(self, query: str, results_count: int = 4, **kwargs) -> RAGWebResults:
+        """
+        Perform a search using the Albert API based on the provided query.
+
+        Args:
+            query (str): The search query.
+            results_count (int): The number of results to return.
+            **kwargs: Additional arguments.
+
+        Returns:
+            RAGWebResults: The search results.
+        """
+        collection_ids = self.get_all_collection_ids()  # might raise RuntimeError
+
+        response = requests.post(
+            urljoin(self._base_url, self._search_endpoint),
+            headers=self._headers,
+            json={
+                "collections": collection_ids,
+                "prompt": query,
+                "score_threshold": 0.6,
+                "k": results_count,  # Number of chunks to return from the search
+            },
+            timeout=settings.ALBERT_API_TIMEOUT,
+        )
+        response.raise_for_status()
+
+        searches = Searches(**response.json())
+
+        return RAGWebResults(
+            data=[
+                RAGWebResult(
+                    url=result.chunk.metadata["document_name"],
+                    content=result.chunk.content,
+                    score=result.score,
+                )
+                for result in searches.data
+            ],
+            usage=RAGWebUsage(
+                prompt_tokens=searches.usage.prompt_tokens,
+                completion_tokens=searches.usage.completion_tokens,
+            ),
+        )
+
+    async def asearch(self, query, results_count: int = 4, **kwargs) -> RAGWebResults:
+        """
+        Perform an asynchronous search using the Albert API based on the provided query.
+
+        Args:
+            query (str): The search query.
+            results_count (int): The number of results to return.
+            **kwargs: Additional arguments.
+
+        Returns:
+            RAGWebResults: The search results.
+        """
+        collection_ids = self.get_all_collection_ids()  # might raise RuntimeError
+
+        async with httpx.AsyncClient(timeout=settings.ALBERT_API_TIMEOUT) as client:
+            response = await client.post(
+                urljoin(self._base_url, self._search_endpoint),
+                headers=self._headers,
+                json={
+                    "collections": collection_ids,
+                    "prompt": query,
+                    "score_threshold": 0.6,
+                    "k": results_count,  # Number of chunks to return from the search
+                },
+                timeout=settings.ALBERT_API_TIMEOUT,
+            )
+
+            logger.debug("Search response: %s %s", response.text, response.status_code)
+
+            response.raise_for_status()
+
+        searches = Searches(**response.json())
+
+        return RAGWebResults(
+            data=[
+                RAGWebResult(
+                    url=result.chunk.metadata["document_name"],
+                    content=result.chunk.content,
+                    score=result.score,
+                )
+                for result in searches.data
+            ],
+            usage=RAGWebUsage(
+                prompt_tokens=searches.usage.prompt_tokens,
+                completion_tokens=searches.usage.completion_tokens,
+            ),
+        )

src/backend/chat/agent_rag/document_rag_backends/base_rag_backend.py

@@ -0,0 +1,210 @@
+"""Implementation of the Albert API for RAG document search."""
+
+import logging
+from abc import ABC, abstractmethod
+from contextlib import asynccontextmanager, contextmanager
+from typing import List, Optional
+
+from asgiref.sync import sync_to_async
+
+from chat.agent_rag.constants import RAGWebResults
+from chat.agent_rag.document_converter.parser import BaseParser
+
+logger = logging.getLogger(__name__)
+
+
+class BaseRagBackend(ABC):
+    """Base class for RAG backends."""
+
+    def __init__(
+        self,
+        collection_id: Optional[str] = None,
+        read_only_collection_id: Optional[List[str]] = None,
+    ):
+        """
+        Backend settings.
+
+        Collection ID is required for RAG operations, where you want to manage the collection
+        lifecycle (create/delete).
+        Read-only collection IDs can be used to access existing collections
+        without managing their lifecycle.
+
+        Collection ID and read-only collection IDs are separated in the implementation to prevent
+        unwanted actions.
+
+        Args:
+            collection_id (Optional[str]): The collection ID for managing the collection lifecycle.
+            read_only_collection_id (Optional[List[str]]): List of read-only collection IDs.
+        """
+        self.collection_id = collection_id
+        self.read_only_collection_id = read_only_collection_id or []
+        self._default_collection_description = "Temporary collection for RAG document search"
+        self.parser: BaseParser = BaseParser()
+
+    @staticmethod
+    def cast_collection_id(collection_id):
+        """Dummy method to be overridden when needed."""
+        return collection_id
+
+    def get_all_collection_ids(self) -> List[str]:
+        """
+        Get all collection IDs, including the main collection ID and read-only collection IDs.
+
+        Returns:
+            List[str]: List of all collection IDs.
+        Raises:
+            RuntimeError: If neither collection_id nor read_only_collection_id is provided.
+        """
+        if not self.collection_id and not self.read_only_collection_id:
+            raise RuntimeError("The RAG backend requires collection_id or read_only_collection_id")
+
+        collection_ids = []
+        if self.collection_id:
+            collection_ids.append(self.cast_collection_id(self.collection_id))
+        if self.read_only_collection_id:
+            collection_ids.extend(
+                [
+                    self.cast_collection_id(collection_id)
+                    for collection_id in self.read_only_collection_id
+                ]
+            )
+        return collection_ids
+
+    @abstractmethod
+    def create_collection(self, name: str, description: Optional[str] = None) -> str:
+        """
+        Create a temporary collection for the search operation.
+        This method should handle the logic to create or retrieve an existing collection.
+        """
+        raise NotImplementedError("Must be implemented in subclass.")
+
+    async def acreate_collection(self, name: str, description: Optional[str] = None) -> str:
+        """
+        Create a temporary collection for the search operation.
+        This method should handle the logic to create or retrieve an existing collection.
+        """
+        return await sync_to_async(self.create_collection)(name=name, description=description)
+
+    def parse_document(self, name: str, content_type: str, content: bytes):
+        """
+        Parse the document and prepare it for the search operation.
+        This method should handle the logic to convert the document
+        into a format suitable for the Albert API.
+
+        Args:
+            name (str): The name of the document.
+            content_type (str): The MIME type of the document (e.g., "application/pdf").
+            content (bytes): The content of the document as a bytes stream.
+
+        Returns:
+            str: The document content in Markdown format.
+        """
+        return self.parser.parse_document(name, content_type, content)
+
+    @abstractmethod
+    def store_document(self, name: str, content: str, **kwargs) -> None:
+        """
+        Store the document content in the collection.
+        This method should handle the logic to send the document content to the API.
+
+        Args:
+            name (str): The name of the document.
+            content (str): The content of the document in Markdown format.
+            **kwargs: Additional arguments. ex: "user_sub" for access control.
+        """
+        raise NotImplementedError("Must be implemented in subclass.")
+
+    async def astore_document(self, name: str, content: str, **kwargs) -> None:
+        """
+        Store the document content in the collection.
+        This method should handle the logic to send the document content to the API.
+
+        Args:
+            name (str): The name of the document.
+            content (str): The content of the document in Markdown format.
+            **kwargs: Additional arguments. ex: "user_sub" for access control.
+        """
+        return await sync_to_async(self.store_document)(name=name, content=content, **kwargs)
+
+    def parse_and_store_document(
+        self, name: str, content_type: str, content: bytes, **kwargs
+    ) -> str:
+        """
+        Parse the document and store it in the Albert collection.
+
+        Args:
+            name (str): The name of the document.
+            content_type (str): The MIME type of the document (e.g., "application/pdf").
+            content (bytes): The content of the document as a bytes stream.
+            **kwargs: Additional arguments. ex: "user_sub" for access control.
+        """
+        if not self.collection_id:
+            raise RuntimeError("The RAG backend requires collection_id")
+
+        document_content = self.parse_document(name, content_type, content)
+        self.store_document(name, document_content, **kwargs)
+        return document_content
+
+    @abstractmethod
+    def delete_collection(self, **kwargs) -> None:
+        """
+        Delete the collection.
+        This method should handle the logic to delete the collection from the backend.
+        """
+        raise NotImplementedError("Must be implemented in subclass.")
+
+    async def adelete_collection(self, **kwargs) -> None:
+        """
+        Delete the collection.
+        This method should handle the logic to delete the collection from the backend.
+        """
+        return await sync_to_async(self.delete_collection)(**kwargs)
+
+    @abstractmethod
+    def search(self, query: str, results_count: int = 4, **kwargs) -> RAGWebResults:
+        """
+        Search the collection for the given query.
+
+        Args:
+            query: The search query string.
+            results_count: Number of results to return.
+            **kwargs: Additional arguments. ex: 'session' for OIDC authentication.
+        """
+        raise NotImplementedError("Must be implemented in subclass.")
+
+    async def asearch(self, query: str, results_count: int = 4, **kwargs) -> RAGWebResults:
+        """
+        Search the collection for the given query asynchronously.
+
+        Args:
+            query: The search query string.
+            results_count: Number of results to return.
+            **kwargs: Additional arguments. ex: 'session' for OIDC authentication.
+        """
+        return await sync_to_async(self.search)(query=query, results_count=results_count, **kwargs)
+
+    @classmethod
+    @contextmanager
+    def temporary_collection(cls, name: str, description: Optional[str] = None):
+        """Context manager for RAG backend with temporary collections."""
+        backend = cls()
+
+        backend.create_collection(name=name, description=description)
+        try:
+            yield backend
+        finally:
+            backend.delete_collection()
+
+    @classmethod
+    @asynccontextmanager
+    async def temporary_collection_async(
+        cls, name: str, description: Optional[str] = None, **kwargs
+    ):
+        """Context manager for RAG backend with temporary collections."""
+        backend = cls()
+
+        await backend.acreate_collection(name=name, description=description)
+        try:
+            yield backend
+        finally:
+            await backend.adelete_collection(**kwargs)

src/backend/chat/agent_rag/document_rag_backends/find_rag_backend.py

@@ -0,0 +1,160 @@
+"""Implementation of the Find API for RAG document search."""
+
+import logging
+import uuid
+from typing import List, Optional
+from urllib.parse import urljoin
+from uuid import uuid4
+
+from django.conf import settings
+from django.utils import timezone
+
+import requests
+
+from chat.agent_rag.constants import RAGWebResult, RAGWebResults, RAGWebUsage
+from chat.agent_rag.document_converter.parser import AlbertParser
+from chat.agent_rag.document_rag_backends.base_rag_backend import BaseRagBackend
+from utils.oidc import with_fresh_access_token
+
+logger = logging.getLogger(__name__)
+
+
+SUPPORTED_LANGUAGE_CODES = ["en", "fr", "de", "nl"]
+
+
+class FindRagBackend(BaseRagBackend):
+    """
+    This class is a placeholder for the Find API implementation.
+    It is designed to be used with the RAG (Retrieval-Augmented Generation) document search system.
+
+    It provides methods to:
+    - Store parsed documents in the Find index.
+    - Perform a search operation using the Find API.
+    """
+
+    def __init__(
+        self,
+        collection_id: Optional[str] = None,
+        read_only_collection_id: Optional[List[str]] = None,
+    ):
+        # Initialize any necessary parameters or configurations here
+        super().__init__(collection_id, read_only_collection_id)
+        self.api_key = settings.FIND_API_KEY
+        self.search_endpoint = "api/v1.0/documents/search/"
+        self.indexing_endpoint = "api/v1.0/documents/index/"
+        self.deleting_endpoint = "api/v1.0/documents/delete/"
+        self.parser = AlbertParser()  # Find Rag relies on Albert parser
+
+    def create_collection(self, name: str, description: Optional[str] = None) -> str:
+        """
+        init collection_id
+        """
+        self.collection_id = self.collection_id or str(uuid.uuid4())
+        return self.collection_id
+
+    @with_fresh_access_token
+    def delete_collection(self, **kwargs) -> None:
+        """
+        Delete the current collection
+        """
+        response = requests.post(
+            urljoin(settings.FIND_API_URL, self.deleting_endpoint),
+            headers={"Authorization": f"Bearer {kwargs['session'].get('oidc_access_token')}"},
+            json={"tags": [f"collection-{self.collection_id}"], "service": "conversations"},
+            timeout=settings.FIND_API_TIMEOUT,
+        )
+        response.raise_for_status()
+
+    def store_document(self, name: str, content: str, **kwargs) -> None:
+        """
+        index document in Find
+
+        Args:
+            name (str): The name of the document.
+            content (str): The content of the document in Markdown format.
+            user_sub (str): The user subject identifier for access control.
+        """
+        logger.debug("index document '%s' in Find", name)
+
+        user_sub = kwargs.get("user_sub")
+        if not user_sub:
+            raise ValueError("user_sub is required to store document in FindRagBackend")
+
+        response = requests.post(
+            urljoin(settings.FIND_API_URL, self.indexing_endpoint),
+            headers={"Authorization": f"Bearer {self.api_key}"},
+            json={
+                "id": str(uuid4()),
+                "title": str(name) or "",
+                "depth": 0,
+                "path": str(name) or "",
+                "numchild": 0,
+                "content": content or "",
+                "created_at": timezone.now().isoformat(),
+                "updated_at": timezone.now().isoformat(),
+                "tags": [f"collection-{self.collection_id}"],
+                "size": len(content.encode("utf-8")),
+                "users": [user_sub],
+                "groups": [],
+                "reach": "authenticated",
+                "is_active": True,
+            },
+            timeout=settings.FIND_API_TIMEOUT,
+        )
+        response.raise_for_status()
+
+    @with_fresh_access_token
+    def search(self, query: str, results_count: int = 4, **kwargs) -> RAGWebResults:
+        """
+        Perform a search using the Find API.
+        Uses the user's OIDC token from the request session.
+
+        Args:
+            query: The search query.
+            results_count: Number of results to return.
+            **kwargs: Additional arguments. Expected: 'session' containing OIDC tokens,
+
+        Returns:
+            RAGWebResults: The search results.
+        """
+        logger.debug("search documents in Find with query '%s'", query)
+        response = requests.post(
+            urljoin(settings.FIND_API_URL, self.search_endpoint),
+            headers={"Authorization": f"Bearer {kwargs['session'].get('oidc_access_token')}"},
+            json={
+                "q": query or "*",
+                "tags": [
+                    f"collection-{collection_id}" for collection_id in self.get_all_collection_ids()
+                ],
+                "k": results_count,
+            },
+            timeout=settings.FIND_API_TIMEOUT,
+        )
+        response.raise_for_status()
+
+        return RAGWebResults(
+            data=[
+                RAGWebResult(
+                    url=get_language_value(result["_source"], "title"),
+                    content=get_language_value(result["_source"], "content"),
+                    score=result["_score"],
+                )
+                for result in response.json()
+            ],
+            usage=RAGWebUsage(
+                prompt_tokens=0,
+                completion_tokens=0,
+            ),
+        )
+
+
+def get_language_value(source, language_field):
+    """
+    extract the value of the language field with the correct language_code extension.
+    "title" and "content" have extensions like "title.en" or "title.fr".
+    get_language_value will return the value regardless of the extension.
+    """
+    for language_code in SUPPORTED_LANGUAGE_CODES:
+        if f"{language_field}.{language_code}" in source:
+            return source[f"{language_field}.{language_code}"]
+    raise ValueError(f"No '{language_field}' field with any supported language code in object")

src/backend/chat/agent_rag/document_search/albert_api.py

@@ -0,0 +1,208 @@
+"""Implementation of the Albert API for RAG document search."""
+
+import json
+import logging
+from io import BytesIO
+from urllib.parse import urljoin
+
+from django.conf import settings
+
+import requests
+
+from chat.agent_rag.albert_api_constants import Searches
+from chat.agent_rag.constants import RAGWebResult, RAGWebResults, RAGWebUsage
+from chat.agent_rag.document_converter.markitdown import DocumentConverter
+from chat.models import ChatConversation
+
+logger = logging.getLogger(__name__)
+
+
+class AlbertRagDocumentSearch:
+    """
+    This class is a placeholder for the Albert API implementation.
+    It is designed to be used with the RAG (Retrieval-Augmented Generation) document search system.
+
+    It provides methods to:
+    - Create a collection for the search operation.
+    - Parse documents and convert them to Markdown format:
+       + Handle PDF parsing using the Albert API.
+       + Use the DocumentConverter (markitdown) for other formats.
+    - Store parsed documents in the Albert collection.
+    - Perform a search operation using the Albert API.
+    """
+
+    def __init__(self, conversation: ChatConversation):
+        # Initialize any necessary parameters or configurations here
+        self._base_url = settings.ALBERT_API_URL
+        self._headers = {
+            "Authorization": f"Bearer {settings.ALBERT_API_KEY}",
+        }
+        self._collections_endpoint = urljoin(self._base_url, "/v1/collections")
+        self._documents_endpoint = urljoin(self._base_url, "/v1/documents")
+        self._pdf_parser_endpoint = urljoin(self._base_url, "/v1/parse-beta")
+        self._search_endpoint = urljoin(self._base_url, "/v1/search")
+
+        self.conversation = conversation
+
+    @property
+    def _albert_collection_id(self):
+        """
+        Generate the collection name based on the conversation ID.
+        This is used to create or retrieve a collection for the search operation.
+        """
+        return f"conversation-{self.conversation.pk}"
+
+    @property
+    def collection_id(self) -> int:
+        """
+        Get the collection ID for the current conversation.
+
+        Might be created later by self._create_collection() if it does not exist.
+        """
+        return int(self.conversation.collection_id) if self.conversation.collection_id else None
+
+    def _create_collection(self) -> bool:
+        """
+        Create a temporary collection for the search operation.
+        This method should handle the logic to create or retrieve an existing collection.
+        """
+        response = requests.post(
+            self._collections_endpoint,
+            headers=self._headers,
+            json={
+                "name": self._albert_collection_id,
+                "description": "Temporary collection for RAG document search",
+                "visibility": "private",
+            },
+            timeout=settings.ALBERT_API_TIMEOUT,
+        )
+        response.raise_for_status()
+        self.conversation.collection_id = str(response.json()["id"])
+        return True
+
+    def _parse_pdf_document(self, name: str, content_type: str, content: BytesIO) -> str:
+        """
+        Parse the PDF document content and return the text content.
+        This method should handle the logic to convert the PDF into
+        a format suitable for the Albert API.
+        """
+        response = requests.post(
+            self._pdf_parser_endpoint,
+            headers=self._headers,
+            files={
+                "file": (
+                    name,
+                    content,
+                    content_type,
+                ),  # Use the name as the filename in the request
+                "output_format": (None, "markdown"),  # Specify the output format as Markdown,
+            },
+            timeout=settings.ALBERT_API_PARSE_TIMEOUT,
+        )
+        response.raise_for_status()
+
+        return "\n\n".join(
+            document_page["content"] for document_page in response.json().get("data", [])
+        )
+
+    def parse_document(self, name: str, content_type: str, content: BytesIO):
+        """
+        Parse the document and prepare it for the search operation.
+        This method should handle the logic to convert the document
+        into a format suitable for the Albert API.
+
+        Args:
+            name (str): The name of the document.
+            content_type (str): The MIME type of the document (e.g., "application/pdf").
+            content (BytesIO): The content of the document as a BytesIO stream.
+
+        Returns:
+            str: The document content in Markdown format.
+        """
+        # Implement the parsing logic here
+        if content_type == "application/pdf":
+            # Handle PDF parsing
+            markdown_content = self._parse_pdf_document(
+                name=name, content_type=content_type, content=content
+            )
+        else:
+            markdown_content = DocumentConverter().convert_raw(
+                name=name, content_type=content_type, content=content
+            )
+
+        return markdown_content
+
+    def _store_document(self, name: str, content: str):
+        """
+        Store the document content in the Albert collection.
+        This method should handle the logic to send the document content to the Albert API.
+
+        Args:
+            content (str): The content of the document in Markdown format.
+        """
+        if not self.collection_id and not self._create_collection():
+            raise RuntimeError("Failed to create or retrieve the collection.")
+
+        response = requests.post(
+            urljoin(self._base_url, self._documents_endpoint),
+            headers=self._headers,
+            files={
+                "file": (f"{name}.md", BytesIO(content.encode("utf-8")), "text/markdown"),
+                "collection": (None, int(self.collection_id)),
+                "metadata": (None, json.dumps({"document_name": name})),  # undocumented API
+            },
+            timeout=settings.ALBERT_API_TIMEOUT,
+        )
+        logger.debug(response.json())
+        response.raise_for_status()
+
+    def parse_and_store_document(self, name: str, content_type: str, content: BytesIO):
+        """
+        Parse the document and store it in the Albert collection.
+
+        Args:
+            name (str): The name of the document.
+            content_type (str): The MIME type of the document (e.g., "application/pdf").
+            content (BytesIO): The content of the document as a BytesIO stream.
+        """
+        document_content = self.parse_document(name, content_type, content)
+        self._store_document(name, document_content)
+        return document_content
+
+    def search(self, query, results_count: int = 4) -> RAGWebResults:
+        """
+        Perform a search using the Albert API based on the provided query.
+
+        :param query: The search query string.
+        :param results_count: The number of results to return.
+        :return: Search results from the Albert API.
+        """
+        response = requests.post(
+            urljoin(self._base_url, self._search_endpoint),
+            headers=self._headers,
+            json={
+                "collections": [self.collection_id],
+                "prompt": query,
+                "score_threshold": 0.6,
+                "k": results_count,  # Number of chunks to return from the search
+            },
+            timeout=settings.ALBERT_API_TIMEOUT,
+        )
+        response.raise_for_status()
+
+        searches = Searches(**response.json())
+
+        return RAGWebResults(
+            data=[
+                RAGWebResult(
+                    url=result.chunk.metadata["document_name"],
+                    content=result.chunk.content,
+                    score=result.score,
+                )
+                for result in searches.data
+            ],
+            usage=RAGWebUsage(
+                prompt_tokens=searches.usage.prompt_tokens,
+                completion_tokens=searches.usage.completion_tokens,
+            ),
+        )

src/backend/chat/agent_rag/web_search/albert_api.py

@@ -0,0 +1,113 @@
+"""
+Albert API web search manager.
+
+Instead of developing a full web search flow, we simply use the Albert API /v1/search endpoint.
+
+See: https://albert.api.etalab.gouv.fr/documentation#tag/Search
+
+Under the hood, on Albert side:
+ - It create a temporary collection
+ - It performs a web search using the query
+ - It returns the results as a list of URls
+ - It loads and parses the content of each URL (vectorization)
+   and stores the results in the temporary collection
+ - It makes a semanctic search on the temporary collection using the query
+ - It returns the results as a list of chunks with metadata
+ - It deletes the temporary collection
+"""
+
+import logging
+from urllib.parse import urljoin
+
+from django.conf import settings
+
+import requests
+
+from ..albert_api_constants import Searches, SearchRequest
+from ..constants import RAGWebResult, RAGWebResults, RAGWebUsage
+from .base import BaseWebSearchManager
+
+logger = logging.getLogger(__name__)
+
+
+class AlbertWebSearchManager(BaseWebSearchManager):
+    """
+    A class to manage web search operations using the Albert API.
+    """
+
+    def __init__(self):
+        """Initializes with the base URL and endpoints for the Albert API."""
+        self._base_url = settings.ALBERT_API_URL
+        self._headers = {
+            "Authorization": f"Bearer {settings.ALBERT_API_KEY}",
+            "Content-Type": "application/json",
+        }
+        self._search_endpoint = urljoin(self._base_url, "/v1/search")
+
+    @staticmethod
+    def _clean_url(url: str) -> str:
+        """
+        Clean the URL by removing the trailing '.html'.
+        We want it to fail when Albert fixes the bug that adds '.html' to the end of URLs.
+        Note: this is a bad workaround because when fixed it may break existing URLs.
+
+        Args:
+            url (str): The URL to clean.
+
+        Returns:
+            str: The cleaned URL.
+        """
+        return url.rsplit(".html", 1)[0]
+
+    def web_search(self, query: str) -> RAGWebResults:
+        """
+        Perform a web search using the Albert API.
+
+        Args:
+            query (str): The search query.
+
+        Returns:
+            Searches: A Searches object containing the search results.
+
+        Raises:
+            ValueError: If the query is empty.
+            requests.HTTPError: If the request to the Albert API fails.
+            requests.exceptions.JSONDecodeError: If the response body does not
+                contain valid json
+        """
+        if not query.strip():
+            raise ValueError("Search query cannot be empty.")
+
+        search_request = SearchRequest(
+            prompt=query,
+            web_search=True,  # Enable web search
+            web_search_k=settings.RAG_WEB_SEARCH_MAX_RESULTS,  # Number of web search results
+            k=settings.RAG_WEB_SEARCH_CHUNK_NUMBER,  # Number of chunks to return from the search
+        )
+
+        logger.debug("Albert API search request: %s", search_request.model_dump())
+
+        response = requests.post(
+            self._search_endpoint,
+            headers=self._headers,
+            json=search_request.model_dump(mode="json", exclude_unset=True),
+            timeout=settings.ALBERT_API_TIMEOUT,
+        )
+        response.raise_for_status()
+
+        searches = Searches(**response.json())
+
+        return RAGWebResults(
+            data=[
+                RAGWebResult(
+                    url=self._clean_url(result.chunk.metadata["document_name"]),
+                    content=result.chunk.content,
+                    score=result.score,
+                )
+                for result in searches.data
+            ],
+            usage=RAGWebUsage(
+                prompt_tokens=searches.usage.prompt_tokens,  # pylint: disable=no-member
+                completion_tokens=searches.usage.completion_tokens,  # pylint: disable=no-member
+            ),
+        )

src/backend/chat/agent_rag/web_search/base.py

@@ -0,0 +1,24 @@
+"""Base class for web search managers."""
+
+from ..constants import RAGWebResults
+
+
+class BaseWebSearchManager:
+    """
+    A class to manage web search operations.
+
+    This is an abstract base class that should be implemented
+    for specific web search managers.
+    """
+
+    def web_search(self, query: str) -> RAGWebResults:
+        """
+        Perform a web search.
+
+        Args:
+            query (str): The search query.
+
+        Returns:
+            RAGWebResults: A Searches object containing the search results.
+        """
+        raise NotImplementedError()

src/backend/chat/agent_rag/web_search/mocked.py

@@ -0,0 +1,68 @@
+"""Mocked web search manager for testing purposes."""
+
+# pylint: disable=line-too-long
+import logging
+
+from ..constants import RAGWebResult, RAGWebResults, RAGWebUsage
+
+logger = logging.getLogger(__name__)
+
+result_1 = {
+    "url": "https://www.lemonde.fr/sciences/article/2025/06/25/le-telescope-james-webb-decouvre-sa-premiere-exoplanete-identifiee-comme-une-petite-planete-froide_6615888_1650684.html",
+    "content": "+ [La beauté créatrice](https://la-beaute-creatrice.lemonde.fr/ \"La beauté créatrice\")\n\t+ [Perspectives](https://www.lemonde.fr/perspectives/ \"Perspectives\")    \n\t+ [Gestion des cookies](#)\n* [Recherche](https://www.lemonde.fr/recherche/)\n\n             Cet article vous est offert Pour lire gratuitement cet article réservé aux abonnés, connectez\\-vous [Se connecter](https://secure.lemonde.fr/sfuser/connexion?gift=true)  Vous n'êtes pas inscrit sur Le Monde ?   \n [Inscrivez\\-vous gratuitement](https://secure.lemonde.fr/sfuser/register?gift=true)        * [Sciences Sciences](https://www.lemonde.fr/sciences/)\n* [Astronomie Astronomie](https://www.lemonde.fr/cosmos/)\n\n  \n\nLe télescope James\\-Webb d��couvre sa première exoplanète, identifiée comme une petite planète froide\n====================================================================================================\n\n   L’observation a été faite grâce à une méthode prometteuse pour détecter des planètes d’une taille similaire à celles du système solaire.     Le Monde avec AFP \n\n  Publié le 25 juin 2025 à 17h49, modifié le 26 juin 2025 à 03h24   Temps de Lecture 2 min. \n\n     Lire plus tard     Partager  * Partager sur Messenger\n* Partager sur Facebook\n* Envoyer par e\\-mail\n* Partager sur Linkedin\n* Copier le lien\n\n         <img src='https://img.lemde.fr/2025/06/25/0/0/866/867/664/0/75/0/1173df3_upload-1-g41rmpa9cg0l-902429.jpg' alt='Une image du disque protoplanétaire autour de l’étoile TWA 7, enregistrée à l’aide de l’instrument Sphere du télescope basé au Chili, est présentée le 25\xa0juin 2025 avec une image superposée capturée par l’instrument MIRI du télescope spatial James-Webb.' title='' width='664' height='443' />   ![Une image du disque protoplanétaire autour de l’étoile TWA 7, enregistrée à l’aide de l’instrument Sphere du télescope basé au Chili, est présentée le 25\xa0juin 2025 avec une image superposée capturée par l’instrument MIRI du télescope spatial James-Webb.](https://img.lemde.fr/2025/06/25/0/0/866/867/664/0/75/0/1173df3_upload-1-g41rmpa9cg0l-902429.jpg)  \n\nUne image du disque protoplanétaire autour de l’étoile TWA 7, enregistrée à l’aide de l’instrument Sphere du télescope basé au Chili, est présentée le 25\xa0juin 2025 avec une image superposée capturée par l’instrument MIRI du télescope spatial James\\-Webb. WST/ESO/REUTERS \n\n   Le télescope spatial James\\-Webb (JWST) a découvert sa première exoplanète dans l’univers proche. Une observation faite grâce à une méthode prometteuse pour détecter des planètes d’une taille similaire à celles du système solaire.\n\n  Depuis 2022, à 1,5\xa0million de kilomètres de la Terre, le JWST a aidé à caractériser plusieurs exoplanètes. *«\xa0Il a passé énormément de temps à observer des planètes qui n’ont jamais été imagées\xa0»*, explique l’astrophysicienne Anne\\-Marie Lagrange, première autrice de l’étude sur le sujet, parue dans [*Nature*](https://www.nature.com/articles/s41550-024-02401-w?utm_source=chatgpt.com \"Nouvelle fenêtre\") mercredi 25\xa0juin.\n\n L’exercice est compliqué du fait que les exoplanètes *«\xa0sont très peu lumineuses parce qu’elles ne sont pas chaudes\xa0»*, mais aussi et surtout du fait qu’*«\xa0on est aveuglé par la lumière de l’étoile autour de laquelle elles tournent\xa0»*, ajoute cette chercheuse du CNRS au Laboratoire d’instrumentation et de recherche en astrophysique de l’Observatoire de Paris.\n\n La parade du James\\-Webb repose sur son coronographe, un instrument qui s’inspire du phénomène de l’éclipse solaire en masquant l’étoile pour mieux révéler ce qui l’entoure, et sur son spectrographe MIRI, capable d’imager les astres les plus discrets grâce à une vision infrarouge. Ses utilisateurs ont pointé le télescope vers l’étoile TWA 7, située dans notre galaxie à une centaine d’années\\-lumière de la Terre, autrement dit sa très petite banlieue. La cible, initialement détectée par le télescope Hubble, était prometteuse à double titre.",
+    "score": 0.51039016,
+}
+result_2 = {
+    "url": "https://www.lemonde.fr/sciences/article/2025/06/25/le-telescope-james-webb-decouvre-sa-premiere-exoplanete-identifiee-comme-une-petite-planete-froide_6615888_1650684.html",
+    "content": "Le télescope James\\-Webb découvre sa première exoplanète, identifiée comme une petite planète froide",
+    "score": 0.49020067,
+}
+result_3 = {
+    "url": "https://www.lemonde.fr/sciences/article/2025/06/25/le-telescope-james-webb-decouvre-sa-premiere-exoplanete-identifiee-comme-une-petite-planete-froide_6615888_1650684.html",
+    "content": "---------------\n\n D’abord, parce qu’elle est jeune de seulement 6,4\xa0millions d’années et donc très susceptible de voir se former des corps planétaires dans le disque de matière la ceinturant. Ensuite, parce que le télescope voit ce disque protoplanétaire par le dessus. Son observation avec l’instrument Sphere du Très Grand Télescope (VLT), situé au Chili, avait permis d’y distinguer trois anneaux s’étageant sur une distance allant jusqu’à plus de cent fois celle séparant la Terre du Soleil.\n\n  Et c’est dans la partie dégarnie du deuxième anneau que l’instrument du James\\-Webb a détecté une *«\xa0source\xa0»* lumineuse, baptisée TWA 7b. Ayant exclu que la découverte s’avère être un objet du système solaire ou une galaxie lointaine, les astronomes l’ont identifiée comme une petite planète froide, d’une masse dix fois inférieure à celles imagées jusqu’ici avec d’autres instruments. Ils estiment sa masse comparable à celle de Saturne, une planète gazeuse qui ne *«\xa0pèse\xa0»* que le tiers de Jupiter, géante gazeuse et poids lourd de notre système solaire.\n\n Avec le James\\-Webb, *«\xa0on est tombé d’un facteur dix en capacité de détection\xa0»*, explique Anne\\-Marie Lagrange, car les planètes les plus *«\xa0légères\xa0»* imagées jusqu’ici depuis le sol pesaient à peu près trois fois la masse de Jupiter. *«\xa0La plupart des autres exoplanètes imagées sont ce qu’on appelle des super\\-Jupiter\xa0»*, ayant de huit à douze fois la masse de cette dernière.\n\n  Lire aussi \\| Article réservé à nos abonnés [Potentielle présence de vie sur l’exoplanète K2\\-18b\xa0: anatomie d’un faux positif](https://www.lemonde.fr/sciences/article/2025/05/13/potentielle-presence-de-vie-sur-l-exoplanete-k2-18b-anatomie-d-un-faux-positif_6605670_1650684.html)      Lire plus tard    \n\nUn télescope prometteur attendu pour 2028",
+    "score": 0.48312354,
+}
+result_4 = {
+    "url": "https://www.lemonde.fr/sciences/article/2025/06/25/le-telescope-james-webb-decouvre-sa-premiere-exoplanete-identifiee-comme-une-petite-planete-froide_6615888_1650684.html",
+    "content": "-----------------------------------------\n\n La performance a d’autant plus d’intérêt que dans le bestiaire planétaire, les planètes rocheuses comme la Terre ou Mars ont des masses beaucoup plus faibles que les planètes gazeuses. Or ces exoplanètes rocheuses constituent une cible ultime des découvreurs de mondes potentiellement habitables.\n\n   Le Monde Guides d’achat   [Aspirateurs robots Les meilleurs aspirateurs robots Lire](https://www.lemonde.fr/guides-d-achat/article/2021/10/25/les-meilleurs-aspirateurs-robots_6099813_5306571.html)   Anne\\-Marie Lagrange souhaiterait désormais *«\xa0découvrir les planètes les plus légères et peut\\-être de trouver des Terres\xa0»*. Avant d’ajouter aussit��t que si *«\xa0on veut comprendre comment les systèmes planétaires se forment, il ne suffit pas de voir les planètes très ou pas massives\xa0»*. Car il faut pouvoir détecter tous les types de planètes, afin de déterminer in fine si notre système solaire est unique ou pas.\n\n Les astronomes estiment que le JWST a le potentiel de détecter et d’imager des planètes ayant une masse encore plus faible que TWA 7b. Mais il faudra de futurs instruments, comme le Télescope extrêmement large (ELT) attendu pour 2028, pour espérer saisir l’image de mondes d’une taille similaire au nôtre.\n\n  Lire aussi \\| Article réservé à nos abonnés [La traque des axions, nouvelle coqueluche des physiciens](https://www.lemonde.fr/sciences/article/2025/06/09/la-traque-des-axions-nouvelle-coqueluche-des-physiciens_6611759_1650684.html)      Lire plus tard      Le Monde avec AFP \n\n     L’espace des contributions est réservé aux abonnés. Abonnez\\-vous pour accéder à cet espace d’échange et contribuer à la discussion.    [S’abonner](https://abo.lemonde.fr/?lmd_medium=BOUTONS_LMFR&lmd_campaign=CONTRIBUTION_ARTICLE)     Contribuer\n\n [Réutiliser ce contenu](https://www.lemonde.fr/syndication/ \"Réutiliser ce contenu\")                                Édition du jour\n\n Daté du jeudi 31 juillet\n\n <img src='data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 276 201'%3E%3C/svg%3E' alt='' title='' width='276' height='201' /> [Lire le journal numérique](https://journal.lemonde.fr/)          [<img src='' alt='' title='' width='' height='36' /> ![]() Culture générale Des leçons interactives par la rédaction pour tester vos connaissances.  Découvrir](https://www.lemonde.fr/memorable/quiz-et-questions-de-culture-generale)         [<img src='' alt='' title='' width='300' height='280' />](https://www.lemonde.fr/chaleur-humaine/article/2025/01/30/mesurez-votre-impact-environnement-avec-le-calculateur-d-empreinte-carbone-et-eau_6523433_6125299.html?lmd_medium=bizdev&lmd_campaign=services_partenaire_lmfr&lmd_creation=ademe)",
+    "score": 0.42882082,
+}
+result_5 = {
+    "url": "https://www.franceinfo.fr/economie/budget/",
+    "content": "Budget 2025 de la France \\- actualité et info en direct\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\ncontenu principal\n\n\n\n\n\n[<img src='https://www.franceinfo.fr/assets/components/headers/header/img/franceinfo-1d7b76a5.svg' alt='' title='' width='161' height='40' />\n\n<img src='data:image/svg+xml,%3Csvg xmlns=\"http://www.w3.org/2000/svg\" viewBox=\"0 0 161 40\"%3E%3C/svg%3E' alt='' title='' width='161' height='40' />\nAccueil France Info](/ \"Accueil France Info\")\n\n* [Rechercher une actualité\nRecherche](/recherche/)\n* [Direct TV\nTV](/en-direct/tv.html)\n* [Direct radio\nRadio](/en-direct/radio.html)\n* [Le live\nLive](/en-direct/)\n* Services\n\n\n\n\n\t+ [Newsletters\n\t (Nouvelle fenêtre)](https://www.francetelevisions.fr/abonnements/information)\n\t+ [Météo\n\t (Nouvelle fenêtre)](https://meteo.franceinfo.fr/)\n\t+ [Jeux\n\t (Nouvelle fenêtre)](https://jeux.franceinfo.fr)\n\t+ [Scolarité](/societe/education/scolarite/)\n* Mon\xa0espace\n\n\n\n\n\t+ [S'inscrire](javascript:void(0))\n\t+ [Se connecter](javascript:void(0))\n\t+ [À lire plus tard](/mon-compte/lire-plus-tard)\n\t+ [Mes commentaires](/mon-compte/interactions)\n\t+ [Mes newsletters](/mon-compte/newsletters)\n\t+ [Mes informations](javascript:void(0))\n\t+ [Se déconnecter](javascript:void(0))\n\n\n\n\n\n\n* [Accueil](/)\n* Menu\n* [Grands formats](/grands-formats/)\n* [Enquêtes](/enquetes-franceinfo/)\n* [Vrai ou faux](/vrai-ou-fake/)\n* [Droits de douane](/monde/usa/droits-de-douane/)\n* [Guerre dans la bande de Gaza](/monde/proche-orient/israel-palestine/)\n* [Loi Duplomb](/environnement/loi-duplomb/)\n\n\n\n\n\n\n\n<img src='https://www.franceinfo.fr/assets/components/headers/header/img/franceinfo-1d7b76a5.svg' alt='France Info' title='' width='129' height='32' />\n\n<img src='data:image/svg+xml,%3Csvg xmlns=\"http://www.w3.org/2000/svg\" viewBox=\"0 0 129 32\"%3E%3C/svg%3E' alt='France Info' title='' width='129' height='32' />\n\n\n\n\n\n\n\n\n\n\n\nRechercher sur franceinfo\n\n\n\n\n\n\nAnnuler la saisie\n\n\n\n\n* [Accueil](/)\n\n\n\n\n En ce moment\n \n\n* [Droits de douane](/monde/usa/droits-de-douane/)\n* [Guerre dans la bande de Gaza](/monde/proche-orient/israel-palestine/)\n* [Loi Duplomb](/environnement/loi-duplomb/)\n\n\n\n\n* [Grands formats](/grands-formats/)\n* [Enquêtes](/enquetes-franceinfo/)\n* [Vrai ou faux](/vrai-ou-fake/)\n\n\n\n\n Rubriques\n \n\n* [L'actu pour les jeunes](/l-actu-pour-les-jeunes/)\n* [Une info transparente](/une-information-transparente-franceinfo/)\n* Monde\n* Europe\n* Faits\\-divers\n* Politique\n* Société\n* Environnement\n* Sport\n* Culture\n* Éco / Conso\n* Santé\n* Sciences \\& technologies\n* Météo\nMétéo\n* Jeux\nJeux\n\n\n\n* Services\n\n\n\nRecevez l'essentiel de l'actualité et restez à jour avec nos newsletters\n\n\n[découvrir nos newsletters\n(Nouvelle fenêtre)](https://www.francetelevisions.fr/abonnements/information)\n\n\n\n Continuez votre exploration\n \n\n* [France 3 régions (nouvelle fenêtre)\nFrance 3 régions](https://france3-regions.franceinfo.fr/)\n* [Outremer la 1ère (nouvelle fenêtre)\nOutremer la 1ère](https://la1ere.franceinfo.fr/)\n* france TV\nfrance TV\n* radio france\nradio france\n\n\n\n\n\n\nServices\n\nServices\n* [Newsletters (nouvelle fenêtre)\nNewsletters](https://www.francetelevisions.fr/abonnements/information)\n* [Météo (nouvelle fenêtre)\nMétéo](https://meteo.franceinfo.fr/)\n* [Jeux (nouvelle fenêtre)\nJeux](https://jeux.franceinfo.fr)\n* [Scolarité](/societe/education/scolarite/)\n\n\n\n\n\n\nMonde\n\nMonde\n\n[voir toute l'actu Monde](/monde/)\n\n\n\n En ce moment\n \n\n* [Droits de douane](/monde/usa/droits-de-douane/)\n* [Guerre dans la bande de Gaza](/monde/proche-orient/israel-palestine/)\n* [Guerre en Ukraine](/monde/europe/manifestations-en-ukraine/)\n* [Affaire Jeffrey Epstein](/monde/usa/affaire-jeffrey-epstein/)\n* [Guerre au Proche\\-Orient](/monde/proche-orient/guerre/)\n* [Donald Trump \\- Président des États\\-Unis](/monde/usa/presidentielle/donald-trump/)\n* [Yémen](/monde/proche-orient/yemen/)\n* [Voir plus de sujets monde](/monde/)\n\n\n\n\n Rubriques",
+    "score": 0.6129482,
+}
+
+
+class MockedWebSearchManager:
+    """
+    A class to manage web search operations.
+
+    This is an abstract base class that should be implemented
+    for specific web search managers.
+    """
+
+    def web_search(self, query: str) -> RAGWebResults:
+        """
+        Fake a web search.
+
+        Args:
+            query (str): The search query.
+
+        Returns:
+            RAGWebResults: A Searches object containing the search results.
+        """
+        logger.info("Mocked web search called with query: %s", query)
+        return RAGWebResults(
+            data=[
+                RAGWebResult(**result_1),
+                RAGWebResult(**result_2),
+                RAGWebResult(**result_3),
+                RAGWebResult(**result_4),
+                RAGWebResult(**result_5),
+            ],
+            usage=RAGWebUsage(
+                prompt_tokens=0,  # Assuming no prompt tokens are provided in this mock
+                completion_tokens=0,  # Assuming no completion tokens are provided in this mock
+            ),  # Assuming no usage data is provided in this mock
+        )

src/backend/chat/agents/base.py

@@ -0,0 +1,190 @@
+"""Base module for PydanticAI agents."""
+
+import dataclasses
+import logging
+
+from django.conf import settings
+from django.core.exceptions import ImproperlyConfigured
+
+import httpx
+from pydantic_ai import Agent
+from pydantic_ai.models import get_user_agent
+from pydantic_ai.profiles import ModelProfile
+
+from chat.tools import get_pydantic_tools_by_name
+
+logger = logging.getLogger(__name__)
+
+
+def prepare_custom_model(configuration: "chat.llm_configuration.LLModel"):
+    """
+    Prepare a custom model instance based on the provided configuration.
+
+    Only few providers are supported at the moment, according to our needs.
+    We define custom models/providers to be able to keep specific configuration
+    when needed.
+    """
+    # pylint: disable=import-outside-toplevel
+
+    match configuration.provider.kind:
+        case "mistral":
+            import pydantic_ai.models.mistral as mistral_models  # noqa: PLC0415
+            from mistralai import TextChunk as MistralTextChunk  # noqa: PLC0415
+            from mistralai import ThinkChunk as MistralThinkChunk  # noqa: PLC0415
+            from mistralai.types.basemodel import Unset as MistralUnset  # noqa: PLC0415
+            from pydantic_ai.providers.mistral import MistralProvider  # noqa: PLC0415
+
+            # --- Monkey patch for pydantic_ai.models.mistral._map_content ---
+            # pylint: disable=protected-access
+
+            # ⚠ WARNING ⚠ WARNING ⚠ WARNING ⚠ WARNING ⚠ WARNING ⚠ WARNING ⚠ WARNING ⚠ WARNING ⚠
+            # |  This workaround is fragile and only works because we are in streaming mode.  |
+            # ⚠ WARNING ⚠ WARNING ⚠ WARNING ⚠ WARNING ⚠ WARNING ⚠ WARNING ⚠ WARNING ⚠ WARNING ⚠
+
+            # The original _map_content raises exceptions for some when responses
+            # contains citation/reference data, which is the case anytime we use
+            # web search or other RAG tool (https://docs.mistral.ai/capabilities/citations/).
+            # We make the patch idempotent using a sentinel attribute so repeated calls
+            # to prepare_custom_model do not re-wrap and do not cause recursive calls.
+            if not getattr(mistral_models, "__safe_map_patched__", False):
+                _original_map_content = mistral_models._map_content  # noqa: SLF001
+
+                def _safe_map_content(content):
+                    """
+                    A safe version of _map_content that ignores unsupported data types.
+
+                    WARNING: this is a monkey patch and may break if the original
+                    function changes in future versions of pydantic_ai.
+                    Current version: pydantic_ai v1.0.18
+                    """
+                    text: str | None = None
+                    thinking: list[str] = []
+
+                    if isinstance(content, MistralUnset) or not content:
+                        return None, []
+
+                    if isinstance(content, list):
+                        for chunk in content:
+                            if isinstance(chunk, MistralTextChunk):
+                                text = (text or "") + chunk.text
+                            elif isinstance(chunk, MistralThinkChunk):
+                                for thought in chunk.thinking:
+                                    if thought.type == "text":  # pragma: no branch
+                                        thinking.append(thought.text)
+                            else:
+                                logger.info(  # pragma: no cover
+                                    "Other data types like (Image, Reference) are not yet "
+                                    "supported,  got %s",
+                                    type(chunk),
+                                )
+                    elif isinstance(content, str):
+                        text = content
+
+                    # Note: Check len to handle potential mismatch between function calls and
+                    # responses from the API.
+                    # (`msg: not the same number of function class and responses`)
+                    if text == "":  # pragma: no cover
+                        text = None
+
+                    return text, thinking
+
+                # Replace the original module-level function
+                mistral_models._map_content = _safe_map_content  # noqa: SLF001
+                mistral_models.__safe_map_patched__ = True
+            # pylint: enable=protected-access
+            # --- End monkey patch ---
+
+            return mistral_models.MistralModel(
+                model_name=configuration.model_name,
+                profile=(
+                    ModelProfile(**configuration.profile.dict(exclude_unset=True))
+                    if configuration.profile
+                    else None
+                ),
+                provider=MistralProvider(
+                    api_key=configuration.provider.api_key,
+                    base_url=configuration.provider.base_url,
+                    # Disable the use of cached client
+                    http_client=httpx.AsyncClient(
+                        timeout=httpx.Timeout(timeout=600, connect=5),
+                        headers={"User-Agent": get_user_agent()},
+                    ),
+                ),
+            )
+        case "openai":
+            from pydantic_ai.models.openai import OpenAIChatModel  # noqa: PLC0415
+            from pydantic_ai.profiles.openai import OpenAIModelProfile  # noqa: PLC0415
+            from pydantic_ai.providers.openai import OpenAIProvider  # noqa: PLC0415
+
+            if configuration.profile and (
+                _config_profile := configuration.profile.dict(exclude_unset=True)
+            ):
+                # set some defaults if not provided, see openai_model_profile which
+                # defines them for known models
+                _model_profile_params = {
+                    "supports_json_schema_output": True,
+                    "supports_json_object_output": True,
+                }
+                _model_profile_params.update(_config_profile)
+                profile = OpenAIModelProfile(**_model_profile_params)
+            else:
+                profile = None
+
+            return OpenAIChatModel(
+                model_name=configuration.model_name,
+                profile=profile,
+                provider=OpenAIProvider(
+                    base_url=configuration.provider.base_url,
+                    api_key=configuration.provider.api_key,
+                ),
+            )
+        case _:
+            raise ImproperlyConfigured(
+                f"Unsupported provider kind '{configuration.provider.kind}' for custom model."
+            )
+
+
+@dataclasses.dataclass(init=False)
+class BaseAgent(Agent):
+    """
+    Base class for PydanticAI agents.
+
+    This class initializes the agent with model from configuration.
+    """
+
+    def __init__(self, *, model_hrid, **kwargs):
+        """Initialize the agent with model configuration from settings."""
+        _ignored_kwargs = {"model", "system_prompt", "tools", "toolsets"}
+        if set(kwargs).intersection(_ignored_kwargs):
+            raise ValueError(f"{_ignored_kwargs} arguments must not be provided.")
+
+        try:
+            self.configuration = settings.LLM_CONFIGURATIONS[model_hrid]
+        except KeyError as exc:
+            raise ImproperlyConfigured(
+                f"LLM model configuration '{model_hrid}' not found."
+            ) from exc
+
+        if self.configuration.is_custom:
+            _model_instance = prepare_custom_model(self.configuration)
+        else:
+            # In this case, we rely on PydanticAI's built-in model registry
+            # and configuration: check pydantic_ai.models.KnownModelName
+            # and pydantic_ai.models.infer_model()
+            _model_instance = self.configuration.model_name
+
+        _system_prompt = self.get_system_prompt()
+
+        _tools = self.get_tools()
+
+        super().__init__(model=_model_instance, instructions=_system_prompt, tools=_tools, **kwargs)
+
+    def get_system_prompt(self) -> str | None:
+        """Override this method to customize the system prompt."""
+        return self.configuration.system_prompt
+
+    def get_tools(self) -> list | None:
+        """Override this method to customize tools."""
+        if not self.configuration.tools:
+            return []
+        return [get_pydantic_tools_by_name(tool_name) for tool_name in self.configuration.tools]

src/backend/chat/agents/conversation.py

@@ -0,0 +1,155 @@
+"""Build the main conversation agent."""
+
+import asyncio
+import dataclasses
+import logging
+
+from django.conf import settings
+from django.utils import formats, timezone
+
+from pydantic_ai import ModelMessage
+from pydantic_ai.models.function import AgentInfo, FunctionModel
+
+from core.enums import get_language_name
+
+from .base import BaseAgent
+
+logger = logging.getLogger(__name__)
+
+MOCKED_RESPONSE = """
+# **Ode to the AI Assistant** 🤖✨
+
+In Paris streets where old meets new, 🗼🇫🇷  
+A helper bright in digital hue,  
+With circuits fast and code so tight,  
+The LaSuite’s bot—oh, what a sight! 🌟
+
+**A chatbot kind**, with wittiness so grand, 💬💡  
+It lends a hand to all the land,  
+From civil servants, bold and wise,  
+To those who seek with hopeful eyes.
+
+It answers quick, it never tires, ⚡🔄  
+With facts and tips to quench desires,  
+A guide so keen, a friend so true,  
+It’s there for **you**—yes, me and you!
+ 
+With **Markdown flair** and emoji cheer, 📝🎨  
+It makes the complex crystal clear,  
+From drafts to code, from sums to prose,  
+It helps the knowledge overflow!  
+
+Oh, **DINUM’s gem**, so sharp, so bright, 💎🌐  
+A beacon in the tech’s vast night,  
+It crafts, it checks, it summarizes,  
+With grace that never compromises.
+
+It **summarizes** the long, the deep, 📚🔍  
+So secrets no more need to sleep,  
+It finds the gems in data’s sea,  
+And sets the truth right there—**for free!**
+
+It **corrects mistakes** with gentle art, ✍️🔄  
+It soothes the mind, it warms the heart,  
+No judgment cast, no frown, no sigh,  
+Just help that’s always standing by.
+
+It **generates code** with swift command, 💻🔥  
+A developer’s dream, first-hand,  
+From Python lines to scripts so neat,  
+It turns the tough to *sweet* and *sweet*!
+
+It **brainstorms ideas**, bold and new, 🧠💡  
+It paints the sky in every hue,  
+From plans to dreams, from start to end,  
+It’s more than code—it’s **trend**, it’s **friend**!
+
+So here’s to you, **Assistant’s pride**, 🏆🎉  
+The bot that’s always by our side,  
+With every prompt, with every line,  
+You make our digital world **divine**!
+
+May you keep shining, bright and true, 🌟🤖  
+The helper every team should woo,  
+For in this age of bits and bytes,  
+You’re **human touch** in tech’s bright lights!
+"""
+
+
+async def mocked_agent_model(_messages: list[ModelMessage], _info: AgentInfo):
+    """
+    Mocked agent model for testing purposes on deployed instances.
+
+    This one only fakes a streamed responses. We could also fake tool calls later.
+    """
+
+    yield "Here is a mocked response (no LLM called)\n---\n"
+    for i in range(0, len(MOCKED_RESPONSE), 4):
+        yield MOCKED_RESPONSE[i : i + 4]
+        await asyncio.sleep(0.03)
+
+
+@dataclasses.dataclass(init=False)
+class ConversationAgent(BaseAgent):
+    """Conversation agent with custom behavior."""
+
+    def __init__(self, *, language=None, **kwargs):
+        """Initialize the conversation agent."""
+        super().__init__(**kwargs)
+
+        # Do not call the real model on deployed instances if the setting is enabled
+        if settings.WARNING_MOCK_CONVERSATION_AGENT:
+            self._model = FunctionModel(stream_function=mocked_agent_model)
+
+        @self.instructions
+        def add_the_date() -> str:
+            """
+            Dynamic instruction function to add the current date.
+
+            Warning: this will always use the date in the server timezone,
+            not the user's timezone...
+            """
+            _formatted_date = formats.date_format(timezone.now(), "l d/m/Y", use_l10n=False)
+            return f"Today is {_formatted_date}."
+
+        @self.instructions
+        def enforce_response_language() -> str:
+            """Dynamic instruction function to set the expected language to use."""
+            return f"Answer in {get_language_name(language).lower()}." if language else ""
+
+    def get_web_search_tool_name(self) -> str | None:
+        """
+        Get the name of the web search tool if available.
+
+        If several are available, return the first one found.
+
+        Warning, this says the tool is available, not that
+        it (the tool/feature) is enabled for the current conversation.
+        """
+        for toolset in self.toolsets:
+            for tool in toolset.tools.values():
+                if tool.name.startswith("web_search_"):
+                    return tool.name
+        return None
+
+
+@dataclasses.dataclass(init=False)
+class TitleGenerationAgent(BaseAgent):
+    """Agent that generates concise, descriptive titles for conversations."""
+
+    def __init__(self, **kwargs):
+        super().__init__(
+            model_hrid=settings.LLM_DEFAULT_MODEL_HRID,
+            output_type=str,
+            **kwargs,
+        )
+
+    def get_tools(self):
+        return []
+
+    def get_system_prompt(self):
+        return (
+            "You are a title generator. Your task is to create concise, descriptive titles "
+            "that accurately summarize conversation content and help the user quickly identify the "
+            "conversation.\n\n"
+        )

src/backend/chat/agents/local_media_url_processors.py

@@ -0,0 +1,187 @@
+"""
+ImageUrl processors and utilities.
+
+Allow to manage local image URLs in messages, replacing them with presigned S3 URLs
+for the LLM to access them, and then reverting them back to local URLs when
+storing the messages in the database.
+"""
+
+import base64
+import logging
+import mimetypes
+import secrets
+from typing import Dict, Iterable
+
+from django.conf import settings
+from django.core.cache import cache
+from django.core.files.storage import default_storage
+
+from pydantic_ai import DocumentUrl, ImageUrl, ModelMessage, ModelRequest, UserPromptPart
+
+from core.file_upload.enums import FileToLLMMode
+from core.file_upload.utils import generate_retrieve_policy
+
+from chat.models import ChatConversation
+
+logger = logging.getLogger(__name__)
+
+
+def generate_temporary_url(key: str) -> str:
+    """
+    Generate a temporary URL for accessing a file through the backend.
+
+    Instead of using S3 presigned URLs, this creates a temporary access key
+    that's stored in cache (3 minutes TTL). The LLM accesses the file through
+    a backend endpoint that validates the key and streams the file content.
+
+    This approach:
+    - Works even when S3 is not accessible from the LLM
+    - Provides better security (key is time-limited and single-use)
+    - Allows the backend to control file access centrally
+
+    Args:
+        key (str): The S3 object key where the file is stored.
+
+    Returns:
+        str: A temporary URL with format: /api/v1.0/file-stream/{temporary_key}/
+    """
+    # Generate a secure random key
+    temporary_key = secrets.token_urlsafe(32)
+
+    # Store the S3 key in cache
+    cache_key = f"file_access:{temporary_key}"
+    cache.set(cache_key, key, timeout=settings.FILE_BACKEND_TEMPORARY_URL_EXPIRATION)
+
+    logger.info("Generated temporary file access key for S3 key: %s", key)
+
+    # Return the URL that the LLM will use to access the file
+    return f"{settings.FILE_BACKEND_URL}/api/v1.0/file-stream/{temporary_key}/"
+
+
+def _get_file_url_for_llm(key: str, mode: str | None = None) -> str:
+    """
+    Get the appropriate URL for the LLM to access a file based on the upload mode.
+
+    Args:
+        key (str): The S3 object key where the file is stored.
+        mode (str, optional): The upload mode. Defaults to FILE_TO_LLM_MODE setting.
+
+    Returns:
+        str: The URL or data URL for the LLM to use.
+
+    Supported modes:
+    - presigned_url: Returns a presigned S3 URL (default)
+    - backend_temporary_url: Returns a presigned URL with shorter expiration
+    - backend_base64: Returns a data URL with base64-encoded file content
+    """
+    if mode is None:
+        mode = settings.FILE_TO_LLM_MODE
+
+    if mode == FileToLLMMode.BACKEND_BASE64:
+        # Read file from S3 and encode as base64 data URL
+        try:
+            with default_storage.open(key, "rb") as file:
+                file_content = file.read()
+                # Detect MIME type from file extension or default to octet-stream
+                mime_type, _ = mimetypes.guess_type(key)
+                if not mime_type:
+                    mime_type = "application/octet-stream"
+
+                # Create data URL
+                b64_content = base64.b64encode(file_content).decode("utf-8")
+                return f"data:{mime_type};base64,{b64_content}"
+        except Exception:  # pylint: disable=broad-except
+            # Fall back to presigned URL on error
+            logger.exception(
+                "Failed to read file for base64 encoding, falling back to presigned URL"
+            )
+            return generate_retrieve_policy(key)
+
+    elif mode == FileToLLMMode.BACKEND_TEMPORARY_URL:
+        return generate_temporary_url(key)
+
+    # FileToLLMMode.PRESIGNED_URL or default
+    return generate_retrieve_policy(key)
+
+
+def update_local_urls(
+    conversation: ChatConversation,
+    contents: Iterable[ImageUrl | DocumentUrl],
+    updated_url: Dict[str, str] | None = None,
+) -> Iterable[ImageUrl | DocumentUrl]:
+    """
+    Replace local image or document URLs in the content list to use appropriate S3 URLs
+    based on the configured FILE_TO_LLM_MODE.
+
+    ⚠️Be careful, `media_contents` are replaced in place.
+
+    Args:
+        conversation (ChatConversation): The chat conversation object.
+        contents (Iterable[ImageUrl | DocumentUrl]): Iterable of UserContent objects.
+        updated_url (Dict[str, str], optional): Dictionary to store
+            mapping of original URLs to updated URLs.
+    Returns:
+        Iterable[ImageUrl | DocumentUrl]: Updated iterable of UserContent objects
+            with appropriate S3 URLs based on the configured mode.
+    """
+    # When images are stored locally, there is no host in the URL, so we can
+    # just check if the URL starts, frontend adds a prefix `/media-key/` to the key.
+    local_media_url_prefix = "/media-key/"
+    local_media_url_prefix_len = len(local_media_url_prefix)
+
+    # Filter only ImageUrl contents
+    media_contents = (c for c in contents if isinstance(c, (ImageUrl, DocumentUrl)))
+
+    # Replace URLs with appropriate S3 URLs based on mode
+    upload_mode = settings.FILE_TO_LLM_MODE
+
+    for content in media_contents:
+        idx = content.url.find(local_media_url_prefix)
+
+        if idx == 0:
+            _initial_url = str(content.url)
+            key = content.url[local_media_url_prefix_len:]
+
+            # Security check: ensure the image belongs to the conversation, if yes,
+            # the user had access to the endpoint, so they have access to the image.
+            if not key.startswith(f"{conversation.pk}/"):
+                # The LLM will throw an error when trying to access the image,
+                # this is not perfect, but this should never happen in practice,
+                # except if the user tampers with the conversation.
+                continue
+
+            content.url = _get_file_url_for_llm(key, upload_mode)
+            if updated_url is not None:
+                updated_url[content.url] = _initial_url
+
+    return contents
+
+
+def update_history_local_urls(
+    conversation: ChatConversation, messages: list[ModelMessage]
+) -> list[ModelMessage]:
+    """
+    Replace local image/documents URLs in the message list to use appropriate S3 URLs.
+
+    ⚠️Be careful, `messages` are replaced in place.
+
+    We don't need to store the mapping of updated URLs to original URLs here because
+    this function is used when sending the history to the LLM (which is already stored
+    in the database with local URLs).
+
+    Args:
+        messages (list[ModelMessage]): List of ModelMessage objects.
+    Returns:
+        list[ModelMessage]: Updated list of ModelMessage objects with appropriate S3 URLs.
+    """
+    # Filter only ModelRequest messages
+    requests = (msg for msg in messages if isinstance(msg, ModelRequest))
+
+    for message in requests:
+        # Filter only UserPromptPart parts
+        user_parts = (part for part in message.parts if isinstance(part, UserPromptPart))
+
+        for part in user_parts:
+            update_local_urls(conversation, part.content)
+
+    return messages

src/backend/chat/agents/summarize.py

@@ -0,0 +1,23 @@
+"""Build the summarization agent."""
+
+import dataclasses
+import logging
+
+from django.conf import settings
+
+from .base import BaseAgent
+
+logger = logging.getLogger(__name__)
+
+
+@dataclasses.dataclass(init=False)
+class SummarizationAgent(BaseAgent):
+    """Create a Pydantic AI summarization Agent instance with the configured settings"""
+
+    def __init__(self, **kwargs):
+        """Initialize the agent with the configured model."""
+        super().__init__(
+            model_hrid=settings.LLM_SUMMARIZATION_MODEL_HRID,
+            output_type=str,
+            **kwargs,
+        )

src/backend/chat/ai_sdk_types.py

@@ -22,7 +22,7 @@ class ToolCall(BaseModel):
 
     toolCallId: str
     toolName: str
-    args: Dict[str, Any]
+    args: Optional[Dict[str, Any]] = None
 
 
 class ToolResult(BaseModel):
@@ -38,7 +38,7 @@ class ToolResult(BaseModel):
 
     toolCallId: str
     toolName: str
-    args: Dict[str, Any]
+    args: Optional[Dict[str, Any]] = None
     result: Any
 
 
@@ -177,15 +177,21 @@ class ToolInvocationUIPart(BaseModel):
 
 class LanguageModelV1Source(BaseModel):
     """
-    Represents source information from a language model.
+    Represents a source that has been used as input to generate the response.
 
     Attributes:
-        source_type: The type of source.
-        details: Additional details about the source.
+        sourceType: A URL source. This is return by web search RAG models.
+        id: The ID of the source.
+        url: The URL of the source.
+        title: The title of the source.
+        providerMetadata: Additional provider metadata for the source.
     """
 
-    source_type: str
-    details: Dict[str, Any]
+    sourceType: Literal["url"]
+    id: str
+    url: str
+    title: Optional[str] = None
+    providerMetadata: Dict[str, Any]  # LanguageModelV1ProviderMetadata
 
 
 class SourceUIPart(BaseModel):

src/backend/chat/clients/async_to_sync.py

@@ -0,0 +1,49 @@
+"""
+Helpers to manage async objects in a synchronous context.
+
+This is not optimal, but we would prefer to stay in a synchronous context
+for now.
+"""
+
+import asyncio
+import queue
+import threading
+
+from chat.clients.exceptions import StreamCancelException
+
+
+def convert_async_generator_to_sync(async_gen):
+    """Convert an async generator to a sync generator."""
+    q = queue.Queue()
+    sentinel = object()
+    exc_sentinel = object()
+
+    async def run_async_gen():
+        try:
+            async for async_item in async_gen:
+                q.put(async_item)
+        except StreamCancelException:
+            # Handle cancellation gracefully, do not put anything in the queue
+            q.put(sentinel)
+        except Exception as exc:  # pylint: disable=broad-except #noqa: BLE001
+            q.put((exc_sentinel, exc))
+        finally:
+            q.put(sentinel)
+
+    def start_async_loop():
+        asyncio.run(run_async_gen())
+
+    thread = threading.Thread(target=start_async_loop, daemon=True)
+    thread.start()
+
+    try:
+        while True:
+            item = q.get()
+            if item is sentinel:
+                break
+            if isinstance(item, tuple) and item[0] is exc_sentinel:
+                # re-raise the exception in the sync context
+                raise item[1]
+            yield item
+    finally:
+        thread.join()

src/backend/chat/clients/exceptions.py

@@ -0,0 +1,17 @@
+"""Module containing custom exceptions for chat clients."""
+
+
+class WebSearchEmptyException(Exception):
+    """Exception raised when a web search returns no results."""
+
+    def __init__(self, message="Web search returned no results."):
+        self.message = message
+        super().__init__(self.message)
+
+
+class StreamCancelException(Exception):
+    """Exception raised when a streaming operation is cancelled."""
+
+    def __init__(self, message="Streaming operation was cancelled."):
+        self.message = message
+        super().__init__(self.message)

src/backend/chat/clients/langchain.py

@@ -1,322 +0,0 @@
-"""LangChainAgentService class for handling AI agent interactions using LangChain."""
-
-import asyncio
-import json
-import logging
-import queue
-import threading
-import uuid
-from contextlib import AsyncExitStack
-from typing import List
-
-from django.conf import settings
-from django.core.exceptions import ImproperlyConfigured
-
-from langchain_core.messages import AIMessageChunk, ToolMessage
-from langchain_core.tools import tool
-
-from chat.ai_sdk_types import (
-    TextUIPart,
-    ToolInvocationPartialCall,
-    ToolInvocationResult,
-    ToolInvocationUIPart,
-    UIMessage,
-)
-from chat.mcp_servers import get_mcp_servers
-
-logger = logging.getLogger(__name__)
-
-# LangChain imports
-from langchain.agents import AgentType, create_react_agent, initialize_agent
-from langchain.chat_models import init_chat_model
-from langchain.schema import (
-    AIMessage,
-    BaseMessage,
-    FunctionMessage,
-    HumanMessage,
-    SystemMessage,
-)
-from langchain.tools import Tool
-
-
-@tool(parse_docstring=True)
-def get_current_weather(location: str, unit: str):
-    """
-    Get the current weather in a given location.
-
-    Args:
-        location (str): The city and state, e.g. San Francisco, CA.
-        unit (str): The unit of temperature, either 'celsius' or 'fahrenheit'.
-    """
-    return {
-        "location": location,
-        "temperature": 22 if unit == "celsius" else 72,
-        "unit": unit,
-    }
-
-
-def convert_async_generator_to_sync(async_gen):
-    """Convert an async generator to a sync generator."""
-    q = queue.Queue()
-    sentinel = object()
-
-    async def run_async_gen():
-        try:
-            async for item in async_gen:
-                q.put(item)
-        finally:
-            q.put(sentinel)
-
-    def start_async_loop():
-        asyncio.run(run_async_gen())
-
-    thread = threading.Thread(target=start_async_loop)
-    thread.start()
-
-    while True:
-        item = q.get()
-        if item is sentinel:
-            break
-        yield item
-
-    thread.join()
-
-
-class AIAgentService:
-    """Service class for AI-related operations using LangChain."""
-
-    def __init__(self, conversation):
-        """Ensure that the AI configuration is set properly."""
-        if settings.AI_API_KEY is None or settings.AI_MODEL is None:
-            raise ImproperlyConfigured("LangChainAgentService configuration not set")
-
-        self.model = init_chat_model(
-            openai_api_key=settings.AI_API_KEY,
-            model=settings.AI_MODEL,
-            model_provider="openai",
-            base_url=settings.AI_BASE_URL,
-        )
-        self.conversation = conversation
-
-    @staticmethod
-    def _convert_to_langchain_messages(messages: List[UIMessage]) -> List[BaseMessage]:
-        lc_messages = []
-        for message in messages:
-            logger.info(f"Converting message: {message}")
-            content = []
-            # Handle main parts
-            for part in message.parts:
-                if part.type == "text":
-                    content.append({"type": "text", "text": part.text})
-                    # content.append(part.text)
-                elif part.type == "tool-invocation":
-                    # Represent tool invocation as a FunctionMessage
-                    tool_invocation = part.toolInvocation
-                    lc_messages.append(
-                        FunctionMessage(
-                            name=tool_invocation.toolName, content=json.dumps(tool_invocation.args)
-                        )
-                    )
-                elif part.type == "image":
-                    # Represent image as a string with a tag or metadata
-                    content.append(
-                        {
-                            "type": "image_url",
-                            "image_url": {"url": part.url},
-                        }
-                    )
-                    # content.append(f"[image: {getattr(part, 'url', '')}]")
-                elif part.type == "file":
-                    # Represent file as a string with a tag or metadata
-                    content.append(
-                        f"[file: {getattr(part, 'name', '')} {getattr(part, 'url', '')}]"
-                    )
-                # Add more types as needed
-
-            # Handle experimental_attachments if present
-            if hasattr(message, "experimental_attachments") and message.experimental_attachments:
-                for attachment in message.experimental_attachments:
-                    if getattr(attachment, "contentType", "").startswith("image"):
-                        content.append(
-                            {
-                                "type": "image_url",
-                                "image_url": {"url": attachment.url},
-                            }
-                        )
-                    elif getattr(attachment, "contentType", "").startswith("text"):
-                        content.append(
-                            f"[file: {getattr(attachment, 'name', '')} {getattr(attachment, 'url', '')}]"
-                        )
-
-            # Compose the message
-            lc_messages.append(
-                {
-                    "role": message.role,
-                    "content": content,
-                }
-            )
-            # if message.role == "system":
-            #    lc_messages.append(SystemMessage(content=content))
-            # elif message.role == "user":
-            #    lc_messages.append(HumanMessage(content=content))
-            # elif message.role == "assistant":
-            #    lc_messages.append(AIMessage(content=content))
-            # FunctionMessage already appended above for tool-invocation
-            # Add more roles/types as needed
-        return lc_messages
-
-    def stream_data(self, messages: List[UIMessage]):  # noqa: PLR0912
-        lc_messages = self._convert_to_langchain_messages(messages)
-        logger.info("[LangChain stream_data_async] Received messages: %s", lc_messages)
-
-        tools = [get_current_weather]
-
-        from langchain_core.prompts import ChatPromptTemplate
-        from langgraph.prebuilt import create_react_agent
-        from langgraph.prebuilt.chat_agent_executor import AgentState
-
-        prompt = ChatPromptTemplate.from_messages(
-            [
-                ("system", "You are a helpful assistant."),
-                ("placeholder", "{messages}"),
-            ]
-        )
-
-        langgraph_agent_executor = create_react_agent(self.model, tools, prompt=prompt)
-
-        finish_reason = "stop"
-
-        _current_tool_call_id = None
-        _current_tool_call_arguments = ""
-        for stream_mode, step in langgraph_agent_executor.stream(
-            {"messages": lc_messages},
-            stream_mode=["messages", "values"],
-        ):
-            # (
-            #     AIMessageChunk(
-            #         content=" with",
-            #         additional_kwargs={},
-            #         response_metadata={},
-            #         id="run--c9fbf725-0ca0-408f-8158-da347a26b64b",
-            #     ),
-            #     {
-            #         "langgraph_step": 1,
-            #         "langgraph_node": "agent",
-            #         "langgraph_triggers": ("branch:to:agent",),
-            #         "langgraph_path": ("__pregel_pull", "agent"),
-            #         "langgraph_checkpoint_ns": "agent:980f9c50-3cfa-6522-f9be-de139f9b4ece",
-            #         "checkpoint_ns": "agent:980f9c50-3cfa-6522-f9be-de139f9b4ece",
-            #         "ls_provider": "openai",
-            #         "ls_model_name": "ai/smollm2",
-            #         "ls_model_type": "chat",
-            #         "ls_temperature": None,
-            #     },
-            # )
-            if stream_mode == "messages":
-                chunk, metadata = step
-
-                try:
-                    logger.info(
-                        "[LangChain stream_data_async] Received chunk: %s %s", type(chunk), chunk
-                    )
-                    if isinstance(chunk, AIMessageChunk):
-                        if chunk.tool_call_chunks:
-                            for tool_call_chunk in chunk.tool_call_chunks:
-                                # Handle tool call chunks
-                                if tool_call_chunk["id"]:
-                                    _current_tool_call_id = tool_call_chunk["id"]
-                                    _tool_name = tool_call_chunk["name"]
-                                    logger.info(
-                                        "[LangChain stream_data_async] Tool call chunk: %s %s",
-                                        _current_tool_call_id,
-                                        _tool_name,
-                                    )
-                                    # yield f'b:{{"toolCallId":"{_current_tool_call_id}","toolName":"{_tool_name}"}}\n'
-
-                                else:
-                                    _argument_delta = tool_call_chunk["args"]
-                                    _current_tool_call_arguments += _argument_delta
-                                    logger.info(
-                                        "[LangChain stream_data_async] Tool call argument delta: %s %s",
-                                        _current_tool_call_id,
-                                        _argument_delta,
-                                    )
-                                    # yield f'c:{{"toolCallId":"{_current_tool_call_id}","argsTextDelta":"{_argument_delta}"}}\n'
-                        elif chunk.response_metadata.get("finish_reason"):
-                            finish_reason = chunk.response_metadata["finish_reason"]
-                            logger.info(
-                                "[LangChain stream_data_async] Finish reason: %s", finish_reason
-                            )
-                        else:
-                            yield f"0:{json.dumps(chunk.content)}\n"
-                    elif isinstance(chunk, ToolMessage) and chunk.content:
-                        _tool_call = {
-                            "toolCallId": chunk.tool_call_id,
-                            "toolName": chunk.name,
-                            "args": json.loads(_current_tool_call_arguments),
-                        }
-                        yield f"9:{json.dumps(_tool_call)}\n"
-                        # content='{"location": "Paris, France", "temperature": 22, "unit": "celsius"}' name='get_current_weather' id='159415d7-046d-43bf-982e-8a69cb50a486' tool_call_id='qk6PXRocvUzh9QTQS6HQNI5qpNujsrzr'
-                        _tool_result_part = {
-                            "toolCallId": chunk.tool_call_id,
-                            "toolName": chunk.name,
-                            "result": json.loads(chunk.content),
-                        }
-                        _current_tool_call_id = None
-                        _current_tool_call_arguments = ""
-                        yield f"a:{json.dumps(_tool_result_part)}\n"
-                    else:
-                        yield f"0:{json.dumps(chunk.content)}\n"
-
-                except Exception as e:
-                    logger.exception("Error in LangChain stream_data_async")
-                    yield f"3:{json.dumps(str(e))}\n"
-                    finish_reason = "error"
-
-            elif stream_mode == "values":
-                logger.info("[LangChain stream_data_async] Received values: %s", step)
-                last_message = step["messages"][-1]
-                # if isinstance(last_message, AIMessage) and last_message.tool_calls:
-                #    for tool_call in last_message.tool_calls:
-                #        _tool_call = {
-                #            "toolCallId": tool_call["id"],
-                #            "toolName": tool_call["name"],
-                #            "args": tool_call["args"],
-                #        }
-                #        yield f"9:{json.dumps(_tool_call)}\n"
-
-        # Simulate finish message
-        _finish_message = {
-            "finishReason": finish_reason,
-            "usage": {},  # LangChain does not provide token usage by default
-        }
-        yield f"d:{json.dumps(_finish_message)}\n"
-
-    def _update_conversation(
-        self,
-        input_messages,
-        result_raw_responses,
-        response_usage,
-    ):
-        # For now, just save the input and output messages
-        # self.conversation.langchain_messages = input_messages
-        self.conversation.messages = self.conversation.ui_messages + [
-            {"content": str(result_raw_responses)}
-        ]
-        self.conversation.save()
-
-    def _convert_langchain_output_to_ui_messages(self, output):
-        # Convert LangChain output to UI messages (simple version)
-        ui_messages = []
-        if isinstance(output, str):
-            text_parts = [TextUIPart(type="text", text=output)]
-            ui_messages.append(
-                UIMessage(
-                    id=str(uuid.uuid4()),
-                    role="assistant",
-                    parts=text_parts,
-                    content=output,
-                )
-            )
-        # Add more conversion logic as needed
-        return ui_messages

src/backend/chat/clients/openai.py

@@ -1,408 +0,0 @@
-"""AIAgentService class for handling AI agent interactions."""
-
-import asyncio
-import json
-import logging
-import queue
-import threading
-import uuid
-from contextlib import AsyncExitStack
-from typing import List
-
-from django.conf import settings
-from django.core.exceptions import ImproperlyConfigured
-
-from agents import Agent, ModelResponse, OpenAIChatCompletionsModel, Runner, Usage
-from asgiref.sync import sync_to_async
-from openai import AsyncOpenAI
-from openai.types.responses import ResponseInputItemParam, ResponseOutputItem
-from openai.types.responses.response_usage import (
-    InputTokensDetails,
-    OutputTokensDetails,
-    ResponseUsage,
-)
-
-from chat.ai_sdk_types import (
-    TextUIPart,
-    ToolInvocationPartialCall,
-    ToolInvocationResult,
-    ToolInvocationUIPart,
-    UIMessage,
-)
-from chat.mcp_servers import get_mcp_servers
-
-logger = logging.getLogger(__name__)
-
-
-def convert_async_generator_to_sync(async_gen):
-    """Convert an async generator to a sync generator."""
-    q = queue.Queue()
-    sentinel = object()
-
-    async def run_async_gen():
-        try:
-            async for item in async_gen:
-                q.put(item)
-        finally:
-            q.put(sentinel)
-
-    def start_async_loop():
-        asyncio.run(run_async_gen())
-
-    thread = threading.Thread(target=start_async_loop)
-    thread.start()
-
-    while True:
-        item = q.get()
-        if item is sentinel:
-            break
-        yield item
-
-    thread.join()
-
-
-class AIAgentService:
-    """Service class for AI-related operations."""
-
-    def __init__(self, conversation):
-        """Ensure that the AI configuration is set properly."""
-        if settings.AI_BASE_URL is None or settings.AI_API_KEY is None or settings.AI_MODEL is None:
-            raise ImproperlyConfigured("AIChatService configuration not set")
-
-        self.model = OpenAIChatCompletionsModel(
-            model=settings.AI_MODEL,
-            openai_client=AsyncOpenAI(
-                base_url=settings.AI_BASE_URL,
-                api_key=settings.AI_API_KEY,
-            ),
-        )
-        self.conversation = conversation
-
-    @staticmethod
-    def _convert_to_openai_messages(
-        messages: List[UIMessage],
-    ) -> List[ResponseInputItemParam]:
-        """Convert UI messages to OpenAI format."""
-        openai_messages = []
-
-        for message in messages:
-            content_parts = []
-            tool_calls = []
-
-            for part in message.parts:
-                match part.type:
-                    case "text":
-                        content_parts.append(
-                            {
-                                "type": "input_text" if message.role == "user" else "output_text",
-                                "text": part.text,
-                            }
-                        )
-
-                    case "image":
-                        content_parts.append(
-                            {
-                                "type": "input_image",
-                                "image_url": part.url,
-                            }
-                        )
-
-                    case "file":
-                        content_parts.append(
-                            {
-                                "type": "input_file",
-                                "file_data": part.url,
-                                "filename": part.name,
-                            }
-                        )
-
-                    case "tool-invocation":
-                        # Extract the tool invocation data
-                        tool_invocation = part.toolInvocation
-                        tool_calls.append(
-                            {
-                                "call_id": tool_invocation.toolCallId,
-                                "type": "function_call",
-                                "name": tool_invocation.toolName,
-                                "arguments": json.dumps(tool_invocation.args),
-                                "status": tool_invocation.state,
-                            }
-                        )
-
-                    case _:
-                        logger.warning("Unrecognized part type: %s in part: %s", part.type, part)
-
-            # Add experimental attachments if they exist
-            if hasattr(message, "experimental_attachments") and message.experimental_attachments:
-                for attachment in message.experimental_attachments:
-                    if attachment.contentType.startswith("image"):
-                        content_parts.append(
-                            {
-                                "type": "input_image",
-                                "image_url": attachment.url,
-                            }
-                        )
-                    elif attachment.contentType.startswith("text"):
-                        content_parts.append(
-                            {
-                                "type": "input_file",
-                                "file_data": attachment.url,
-                                "filename": attachment.name,
-                            }
-                        )
-
-            # Add message with content parts if there are any
-            if content_parts:
-                openai_messages.append(
-                    {"role": message.role, "content": content_parts, "type": "message"}
-                )
-
-            # Add tool calls separately, will be merged by `items_to_messages`
-            openai_messages += tool_calls
-
-        return openai_messages
-
-    def stream_text(self, messages: List[UIMessage]):
-        """Simple generator to convert async generator to sync generator."""
-        async_generator = self.stream_text_async(messages)
-        return convert_async_generator_to_sync(async_generator)
-
-    async def stream_text_async(self, messages: List[UIMessage]):
-        """Async generator for streaming agent events."""
-        openai_messages = self._convert_to_openai_messages(messages)
-        logger.debug("[stream_data_async] Received messages: %s", openai_messages)
-
-        async with AsyncExitStack() as stack:
-            initialized_mcp_servers = [
-                await stack.enter_async_context(mcp_server) for mcp_server in get_mcp_servers()
-            ]
-            agent = Agent(
-                name=settings.AI_AGENT_NAME,
-                instructions=settings.settings.AI_AGENT_INSTRUCTIONS,
-                model=self.model,
-                #tools=[agent_get_current_weather],
-                mcp_servers=initialized_mcp_servers,
-            )
-
-            result = Runner.run_streamed(
-                agent,
-                input=openai_messages,
-            )
-
-            async for event in result.stream_events():
-                logger.debug("[stream_text_async] Received event: %s", event)
-                if event.type == "raw_response_event":
-                    data = event.data
-                    logger.debug("[stream_text_async]  - data: %s", data)
-                    if data.type == "response.output_text.delta":
-                        yield data.delta
-
-            # At the end, save the response and yield the finish message part
-            _response_usage = Usage()
-            for raw_response in result.raw_responses:
-                _response_usage.add(raw_response.usage)
-
-            await sync_to_async(self._update_conversation)(
-                openai_messages, result.raw_responses, _response_usage
-            )
-
-    def stream_data(self, messages: List[UIMessage]):
-        """Simple generator to convert async generator to sync generator."""
-        async_generator = self.stream_data_async(messages)
-        return convert_async_generator_to_sync(async_generator)
-
-    async def stream_data_async(self, messages: List[UIMessage]):
-        """Async generator for streaming agent events."""
-        finish_reason = "stop"
-
-        openai_messages = self._convert_to_openai_messages(messages)
-        logger.debug("[stream_data_async] Received messages: %s", openai_messages)
-
-        async with AsyncExitStack() as stack:
-            initialized_mcp_servers = [
-                await stack.enter_async_context(mcp_server) for mcp_server in get_mcp_servers()
-            ]
-
-            # websearch_agent = Agent(
-            #     name="web search",
-            #     instructions=(
-            #         "You are a web search agent. "
-            #         "Your task is to search the web for up-to-date information."
-            #         " You will be called by the main agent to perform web searches."
-            #         " You will receive a query and return the search results."
-            #         " The results should be in the format of a list of dictionaries, "
-            #         "each containing 'link', 'title', and 'snippet' keys."
-            #         " If you cannot find any results, return an empty list."
-            #         " Do not include any other information in your response."
-            #         " Do not include any additional text or explanations."
-            #         " You must always annotate the response mentioning the url, title, and snippet."
-            #     ),
-            #     handoff_description="You are a web search agent. Your task is to search the web for up-to-date information.",
-            #     model=self.model,
-            #     tools=[agent_web_search_tavily],
-            # )
-
-            agent = Agent(
-                name=settings.AI_AGENT_NAME,
-                instructions=settings.AI_AGENT_INSTRUCTIONS,
-                model=self.model,
-                mcp_servers=initialized_mcp_servers,
-                # handoffs=[websearch_agent],
-            )
-            result = Runner.run_streamed(
-                agent,
-                input=openai_messages,
-            )
-
-            try:
-                async for event in result.stream_events():
-                    logger.debug("[stream_data_async] Received event: %s", event)
-
-                    if event.type == "raw_response_event":
-                        data = event.data
-                        if data.type == "response.output_text.delta":
-                            yield f"0:{json.dumps(data.delta)}\n"
-
-                        if hasattr(data, "finish_reason") and data.finish_reason:
-                            finish_reason = data.finish_reason
-
-                    elif event.type == "run_item_stream_event":
-                        item = event.item
-                        if item.type == "tool_call_item":
-                            _tool_call = {
-                                "toolCallId": item.raw_item.call_id,
-                                "toolName": item.raw_item.name,
-                                "args": (
-                                    json.loads(item.raw_item.arguments)
-                                    if hasattr(item.raw_item, "arguments")
-                                    else {}
-                                ),
-                            }
-                            yield f"9:{json.dumps(_tool_call)}\n"
-                        elif item.type == "tool_call_output_item":
-                            _tool_call_result = {
-                                "toolCallId": str(item.raw_item["call_id"]),
-                                "result": item.raw_item["output"],
-                            }
-                            yield f"a:{json.dumps(_tool_call_result)}\n"
-                    elif event.type == "agent_updated_stream_event":
-                        logger.debug(
-                            "[stream_data_async] Agent switched to: %s", event.new_agent.name
-                        )
-
-            except Exception as e:  # pylint: disable=broad-except
-                logger.exception("Error in stream_data_async")
-                yield f"3:{json.dumps(str(e))}\n"
-                finish_reason = "error"
-
-            # At the end, save the response and yield the finish message part
-            _response_usage = Usage()
-            for raw_response in result.raw_responses:
-                _response_usage.add(raw_response.usage)
-
-            await sync_to_async(self._update_conversation)(
-                openai_messages, result.raw_responses, _response_usage
-            )
-
-            _finish_message = {
-                "finishReason": finish_reason,
-                "usage": {
-                    "promptTokens": _response_usage.input_tokens,
-                    "completionTokens": _response_usage.output_tokens,
-                },
-            }
-            yield f"d:{json.dumps(_finish_message)}\n"
-
-    def _update_conversation(
-        self,
-        input_messages: List[ResponseInputItemParam],
-        result_raw_responses: List[ModelResponse],
-        response_usage: Usage,
-    ):
-        ui_messages = []
-
-        self.conversation.openai_messages = input_messages + [
-            output.model_dump()
-            for raw_response in result_raw_responses
-            for output in raw_response.output
-        ]
-
-        for raw_response in result_raw_responses:
-            ui_messages += self._convert_openai_output_to_ui_messages(raw_response.output)
-
-        self.conversation.messages = self.conversation.ui_messages + [
-            ui_message.model_dump() for ui_message in ui_messages
-        ]
-
-        if self.conversation.agent_usage:
-            total_usage = ResponseUsage(**self.conversation.agent_usage)
-        else:
-            total_usage = ResponseUsage(
-                input_tokens=0,
-                output_tokens=0,
-                input_tokens_details=InputTokensDetails(cached_tokens=0),
-                output_tokens_details=OutputTokensDetails(reasoning_tokens=0),
-                total_tokens=0,
-            )
-
-        total_usage.input_tokens += response_usage.input_tokens  # pylint: disable=no-member
-        total_usage.output_tokens += response_usage.output_tokens  # pylint: disable=no-member
-        total_usage.input_tokens_details.cached_tokens += (
-            response_usage.input_tokens_details.cached_tokens
-        )
-        total_usage.output_tokens_details.reasoning_tokens += (
-            response_usage.output_tokens_details.reasoning_tokens
-        )
-        total_usage.total_tokens = response_usage.total_tokens
-
-        self.conversation.agent_usage = total_usage.model_dump()
-
-        self.conversation.save()
-
-    def _convert_openai_output_to_ui_messages(
-        self, output: List[ResponseOutputItem]
-    ) -> List[UIMessage]:
-        """Convert OpenAI output to UI messages."""
-        ui_messages = []
-
-        for item in output:
-            if item.type == "message":
-                text_parts = [TextUIPart(type="text", text=item.text) for item in item.content]
-                ui_messages.append(
-                    UIMessage(
-                        id=str(uuid.uuid4()),
-                        role="assistant",
-                        parts=text_parts,
-                        content="".join(part.text for part in text_parts),
-                    )
-                )
-            elif item.type == "function_call":
-                if item.status == "in_progress":
-                    tool_invocation = ToolInvocationPartialCall(
-                        state="partial-call",
-                        step=None,
-                        toolCallId=item.call_id,
-                        toolName=item.name,
-                        args=json.loads(item.arguments),
-                    )
-                elif item.status == "completed":
-                    tool_invocation = ToolInvocationResult(
-                        state="result",
-                        step=None,
-                        toolCallId=item.call_id,
-                        toolName=item.name,
-                        args=json.loads(item.arguments),
-                        result=json.loads(item.result) if item.result else None,
-                    )
-                # elif item.status == "incomplete":
-                else:
-                    logger.warning("[stream_data_async] Unhandled message: %s", item)
-                    continue
-
-                ui_tool_invocation = ToolInvocationUIPart(
-                    type="tool-invocation",
-                    toolInvocation=tool_invocation,
-                )
-                ui_messages.append(UIMessage(role="assistant", parts=[ui_tool_invocation]))
-            # Handle other types as needed
-        return ui_messages

src/backend/chat/clients/pydantic_ui_message_converter.py

@@ -0,0 +1,252 @@
+"""
+Utility functions to convert between UIMessage (ai_sdk_types.py)
+and UserContent/ModelMessage (pydantic_ai.messages.py).
+"""
+
+import base64
+import json
+import logging
+import uuid
+from dataclasses import asdict
+from typing import List
+
+from pydantic_ai.messages import (
+    BinaryContent,
+    DocumentUrl,
+    ImageUrl,
+    ModelMessage,
+    ModelRequest,
+    ModelResponse,
+    RetryPromptPart,
+    SystemPromptPart,
+    TextPart,
+    ThinkingPart,
+    ToolCallPart,
+    ToolReturnPart,
+    UserContent,
+    UserPromptPart,
+)
+
+from chat.ai_sdk_types import (
+    Attachment,
+    FileUIPart,
+    ReasoningDetailText,
+    ReasoningUIPart,
+    TextUIPart,
+    ToolInvocationCall,
+    ToolInvocationUIPart,
+    UIMessage,
+    UIPart,
+)
+
+
+def ui_message_to_user_content(message: UIMessage) -> List[UserContent]:
+    """
+    Convert a UIMessage to a list of UserContent for Pydantic-AI.
+    """
+    user_contents: List[UserContent] = []
+    for part in message.parts:
+        if isinstance(part, TextUIPart):
+            user_contents.append(part.text)
+        elif isinstance(part, FileUIPart):
+            user_contents.append(
+                BinaryContent(data=part.data.encode("utf-8"), media_type=part.mimeType)
+            )
+        elif isinstance(part, ToolInvocationUIPart):
+            # Tool invocations are not directly mapped to UserContent, skip or handle as needed
+            continue
+        elif isinstance(part, ReasoningUIPart):
+            # Reasoning parts are not directly mapped to UserContent, skip or handle as needed
+            continue
+        else:
+            raise ValueError(f"Unsupported UIPart type: {type(part)}")
+    for experimental_attachment in message.experimental_attachments or []:
+        if experimental_attachment.url.startswith("data:"):
+            # Handle data URLs
+            raw_data = base64.b64decode(experimental_attachment.url.split(",")[1])
+            user_contents.append(
+                BinaryContent(
+                    data=raw_data,
+                    media_type=experimental_attachment.contentType,
+                    identifier=experimental_attachment.name,
+                )
+            )
+        elif experimental_attachment.contentType.startswith("image/"):
+            user_contents.append(
+                ImageUrl(
+                    url=experimental_attachment.url,
+                    media_type=experimental_attachment.contentType,
+                    identifier=experimental_attachment.name,
+                )
+            )
+        else:
+            user_contents.append(
+                DocumentUrl(
+                    url=experimental_attachment.url,
+                    media_type=experimental_attachment.contentType,
+                    identifier=experimental_attachment.name,
+                )
+            )
+
+    return user_contents
+
+
+def model_message_to_ui_message(model_message: ModelMessage) -> UIMessage:  # noqa: PLR0912, PLR0915  # pylint: disable=too-many-statements
+    """
+    Convert a ModelMessage (ModelRequest or ModelResponse) to a UIMessage.
+    """
+    # pylint: disable=too-many-nested-blocks,too-many-branches
+    parts: List[UIPart] = []
+    experimental_attachments: List[Attachment] = []
+
+    logging.getLogger(__name__).debug(
+        "Converting ModelMessage to UIMessage: %s %s",
+        type(model_message),
+        asdict(model_message),
+    )
+    _states = {"tool-calls": {}}
+
+    if isinstance(model_message, ModelRequest):
+        message_timestamp = None
+
+        for part in model_message.parts:
+            if isinstance(part, SystemPromptPart):
+                # System prompts are not included in UIMessage parts
+                continue
+            if isinstance(part, UserPromptPart):
+                message_timestamp = part.timestamp
+                if isinstance(part.content, str):
+                    parts.append(TextUIPart(type="text", text=part.content))
+                elif isinstance(part.content, list):
+                    for c in part.content:
+                        if isinstance(c, str):
+                            parts.append(TextUIPart(type="text", text=c))
+                        elif isinstance(c, BinaryContent):
+                            experimental_attachments.append(
+                                Attachment(
+                                    contentType=c.media_type,
+                                    url=f"data:{c.media_type};base64,"
+                                    + base64.b64encode(c.data).decode("utf-8"),
+                                )
+                            )
+                        elif isinstance(c, ImageUrl):
+                            experimental_attachments.append(
+                                Attachment(
+                                    contentType=c.media_type,
+                                    url=c.url,
+                                    name=c.identifier,
+                                )
+                            )
+                        elif isinstance(c, DocumentUrl):
+                            experimental_attachments.append(
+                                Attachment(
+                                    contentType=c.media_type,
+                                    url=c.url,
+                                    name=c.identifier,
+                                )
+                            )
+                        else:  # AudioUrl, VideoUrl
+                            raise ValueError(
+                                f"Unsupported UserContent in UserPromptPart: {type(c)}"
+                            )
+            elif isinstance(part, TextPart) and part.content:
+                parts.append(TextUIPart(type="text", text=part.content))
+            elif isinstance(part, ToolReturnPart):
+                pass
+                # parts.append(ToolInvocationUIPart(
+                #     type="tool-invocation",
+                #     toolInvocation=ToolInvocationResult(
+                #         state="result",
+                #         toolCallId=part.tool_call_id,
+                #         toolName=part.tool_name,
+                #         args={},
+                #         result=part.content,
+                #     )
+                # ))
+            elif isinstance(part, ThinkingPart):
+                parts.append(
+                    ReasoningUIPart(
+                        type="reasoning",
+                        reasoning=part.content,
+                        details=[
+                            ReasoningDetailText(
+                                type="text",
+                                text=part.content,
+                                signature=part.signature,
+                            )
+                        ],
+                    )
+                )
+            elif isinstance(part, RetryPromptPart):
+                # Retry prompts are not included in UIMessage parts
+                continue
+            else:
+                raise ValueError(f"Unsupported ModelRequest part type: {type(part)}")
+
+        if not parts:
+            return None
+
+        return UIMessage(
+            id=str(uuid.uuid4()),
+            role="user",
+            content="".join(part.text for part in parts if isinstance(part, TextUIPart)),
+            parts=parts,
+            experimental_attachments=experimental_attachments or None,
+            createdAt=message_timestamp,
+        )
+
+    if isinstance(model_message, ModelResponse):
+        for part in model_message.parts:
+            if isinstance(part, UserPromptPart):
+                if isinstance(part.content, str):
+                    parts.append(TextUIPart(type="text", text=part.content))
+                elif isinstance(part.content, list):
+                    for c in part.content:
+                        if isinstance(c, str):
+                            parts.append(TextUIPart(type="text", text=c))
+                        else:  # ImageUrl, AudioUrl, VideoUrl, DocumentUrl, BinaryContent
+                            raise ValueError(
+                                f"Unsupported UserContent in UserPromptPart: {type(c)}"
+                            )
+            elif isinstance(part, TextPart):
+                parts.append(TextUIPart(type="text", text=part.content))
+            elif isinstance(part, ToolCallPart):
+                parts.append(
+                    ToolInvocationUIPart(
+                        type="tool-invocation",
+                        toolInvocation=ToolInvocationCall(
+                            state="call",
+                            toolCallId=part.tool_call_id,
+                            toolName=part.tool_name,
+                            args=json.loads(part.args)
+                            if isinstance(part.args, str)
+                            else part.args or {},
+                        ),
+                    )
+                )
+            elif isinstance(part, ThinkingPart):
+                parts.append(
+                    ReasoningUIPart(
+                        type="reasoning",
+                        reasoning=part.content,
+                        details=[
+                            ReasoningDetailText(
+                                type="text",
+                                text=part.content,
+                                signature=part.signature,
+                            )
+                        ],
+                    )
+                )
+            else:
+                raise ValueError(f"Unsupported ModelMessage part type: {type(part)}")
+
+        return UIMessage(
+            id=str(uuid.uuid4()),
+            role="assistant",
+            content="".join(part.text for part in parts if isinstance(part, TextUIPart)),
+            parts=parts,
+            createdAt=model_message.timestamp,
+        )
+
+    raise ValueError(f"Unsupported ModelMessage part type: {type(model_message)}")

src/backend/chat/factories.py

@@ -0,0 +1,56 @@
+"""Factories for chat application."""
+
+from uuid import uuid4
+
+import factory.django
+
+from core.factories import UserFactory
+
+from . import models
+
+
+class ChatProjectFactory(factory.django.DjangoModelFactory):
+    """Factory for creating Project instances."""
+
+    title = factory.Sequence(lambda n: f"title {n}")
+    owner = factory.SubFactory(UserFactory)
+    icon = factory.fuzzy.FuzzyChoice(models.ChatProjectIcon)
+    color = factory.fuzzy.FuzzyChoice(models.ChatProjectColor)
+
+    class Meta:
+        model = models.ChatProject
+        skip_postgeneration_save = True
+
+    @factory.post_generation
+    def number_of_conversations(self, create, extracted, **kwargs):
+        """Create attached conversations for the project."""
+        if not create or not extracted:
+            return
+
+        if not isinstance(extracted, int):
+            raise TypeError("number_of_conversations must be an integer")
+        ChatConversationFactory.create_batch(extracted, project=self, owner=self.owner)
+
+
+class ChatConversationFactory(factory.django.DjangoModelFactory):
+    """Factory for creating ChatConversation instances."""
+
+    owner = factory.SubFactory(UserFactory)
+
+    class Meta:
+        model = models.ChatConversation
+
+
+class ChatConversationAttachmentFactory(factory.django.DjangoModelFactory):
+    """Factory for creating ChatConversationAttachment instances."""
+
+    conversation = factory.SubFactory(ChatConversationFactory)
+    uploaded_by = factory.SubFactory(UserFactory)
+    key = factory.LazyAttribute(
+        lambda obj: f"{obj.conversation.pk}/attachments/{uuid4()}.{obj.file_name.split('.')[-1]}"
+    )
+    file_name = factory.Faker("file_name")
+    content_type = factory.Faker("mime_type")
+
+    class Meta:
+        model = models.ChatConversationAttachment

src/backend/chat/keepalive.py

@@ -0,0 +1,171 @@
+"""Helpers to prevent proxy timeouts during long-running stream operations.
+
+This module provides utilities to wrap synchronous and asynchronous iterators
+with keepalive messages. When a stream pauses for longer than the specified
+interval, keepalive messages are injected to prevent proxy/gateway
+timeouts while waiting for the stream data.
+"""
+
+import asyncio
+import logging
+import queue
+import threading
+import time
+from typing import AsyncIterator, Iterator
+
+from django.conf import settings
+
+from .vercel_ai_sdk.core.events_v4 import DataPart as V4DataPart
+from .vercel_ai_sdk.core.events_v5 import DataPart as V5DataPart
+from .vercel_ai_sdk.encoder import (
+    CURRENT_EVENT_ENCODER_VERSION,
+    EventEncoder,
+    EventEncoderVersion,
+)
+
+logger = logging.getLogger(__name__)
+
+
+def get_keepalive_message() -> str:
+    """Generate a keepalive message based on encoder/SDK version."""
+    if CURRENT_EVENT_ENCODER_VERSION == EventEncoderVersion.V4:
+        event = V4DataPart(data=[{"status": "WAITING"}])
+    else:
+        event = V5DataPart(data={"status": "WAITING"})
+    encoder = EventEncoder(CURRENT_EVENT_ENCODER_VERSION)
+    return encoder.encode(event)
+
+
+async def stream_with_keepalive_async(
+    stream: AsyncIterator[str],
+) -> AsyncIterator[str]:
+    """Wrap an async iterator to emit keepalive during long pauses.
+
+    Args:
+        stream: The async iterator to wrap
+    Yields:
+        Items from the original stream, plus keepalive messages during pauses
+    Raises:
+        Any exception raised by the original stream
+    """
+    q: asyncio.Queue = asyncio.Queue()
+    finished = asyncio.Event()
+    keepalive_message = get_keepalive_message()
+
+    async def producer():
+        """Background task that consumes the original stream into a queue."""
+
+        try:
+            async for stream_item in stream:
+                await q.put(stream_item)
+        except Exception as exc:  # pylint: disable=broad-except #noqa: BLE001
+            # Pass exceptions through the queue so the consumer can re-raise them.
+            # This ensures errors aren't silently swallowed.
+            await q.put(exc)
+        finally:
+            finished.set()
+            await q.put(None)  # Sentinel to signal completion
+
+    producer_task = asyncio.create_task(producer())
+
+    try:
+        while True:
+            try:
+                item = await asyncio.wait_for(q.get(), timeout=settings.KEEPALIVE_INTERVAL)
+                if item is None:
+                    break
+                if isinstance(item, Exception):
+                    raise item
+                yield item
+            except asyncio.TimeoutError:
+                # No data received within interval
+                if finished.is_set():
+                    # Producer is done, queue is empty (else we would not have timed out)
+                    break
+
+                logger.debug("Send keepalive")
+                yield keepalive_message
+    finally:
+        # Cleanup
+        producer_task.cancel()
+        try:
+            await producer_task
+        except asyncio.CancelledError:
+            pass
+
+
+def get_current_time() -> float:
+    """Get current monotonic time, avoiding freezegun interferences.
+
+    Returns time.monotonic() which:
+    - Is NOT affected by freezegun's @freeze_time decorator (unlike time.time())
+    - Prevents issues where frozen time in main thread differs from real time in
+      spawned threads, causing incorrect keepalive interval computation
+    - Is the best clock for measuring time intervals
+
+    Wrapped in a function to ease mocking in tests.
+
+    Returns:
+        float: Monotonic time in seconds since an arbitrary reference point
+    """
+    return time.monotonic()
+
+
+def stream_with_keepalive_sync(stream: Iterator[str]) -> Iterator[str]:
+    """Wraps a synchronous stream with keepalive messages."""
+
+    q: queue.Queue = queue.Queue()
+    stream_done = threading.Event()
+    keepalive_message = get_keepalive_message()
+    # Mutable container so threads can read/write shared timestamp
+    last_yield_time = [get_current_time()]
+
+    def consume_stream():
+        """Read from source stream and forward chunks to queue."""
+        try:
+            for chunk in stream:
+                if stream_done.is_set():
+                    return  # early exit
+                q.put(chunk, timeout=1)  # Arbitrary timeout prevents blocking forever
+        # pylint: disable=broad-exception-caught
+        except Exception as e:
+            logger.exception("Error in stream consumption")
+            q.put(e)
+        finally:
+            stream_done.set()
+
+    def send_keepalives():
+        """Inject keepalive messages when idle too long.
+
+        Uses get_current_time() (time.monotonic) instead of time.time()
+        to avoid issues with freezegun in tests.
+        """
+        while not stream_done.is_set():
+            # Sleep before checking to give main loop time to process and update timestamp
+            time.sleep(0.5)  # let main loop process first, empiric value
+            if get_current_time() - last_yield_time[0] >= settings.KEEPALIVE_INTERVAL:
+                try:
+                    q.put(keepalive_message, timeout=0.1)
+                except queue.Full:
+                    pass
+
+    for target in (consume_stream, send_keepalives):
+        threading.Thread(target=target, daemon=True).start()
+
+    try:
+        # Continue while stream is active or queue has still items
+        while not stream_done.is_set() or not q.empty():
+            try:
+                item = q.get(timeout=1)  # short timeout, avoid blocking and stay responsive
+            except queue.Empty:
+                continue
+
+            # Re-raise from consume_stream
+            if isinstance(item, Exception):
+                raise item
+
+            yield item
+            last_yield_time[0] = get_current_time()
+    finally:
+        # Signal threads to stop
+        stream_done.set()

src/backend/chat/llm_configuration.py

@@ -0,0 +1,198 @@
+"""Module for managing LLM configurations from a JSON configuration file."""
+
+import os
+from functools import lru_cache
+from typing import Annotated, Any, Literal, Optional, Self, Sequence
+
+from pydantic import (
+    AfterValidator,
+    BaseModel,
+    BeforeValidator,
+    Field,
+    ImportString,
+    field_validator,
+    model_validator,
+)
+from pydantic_ai.profiles import JsonSchemaTransformer
+
+
+def _get_setting_or_env_or_value(value: str) -> Any:
+    """Get the value from environment variable, Django settings, or return the value as is."""
+    from django.conf import settings  # pylint: disable=import-outside-toplevel # noqa: PLC0415
+
+    if value.startswith("environ."):
+        env_var = value.split("environ.")[1]
+        new_value = os.environ.get(env_var, None)
+        if new_value is None:
+            raise ValueError(f"Environment variable '{env_var}' not set.")
+        return new_value
+
+    if value.startswith("settings."):
+        setting_var = value.split("settings.")[1]
+        new_value = getattr(settings, setting_var, None)
+        if new_value is None:
+            raise ValueError(f"Django setting '{setting_var}' not set.")
+        return new_value
+
+    return value
+
+
+SettingEnvValue = Annotated[
+    str,
+    AfterValidator(_get_setting_or_env_or_value),
+]
+
+LongStringAsListValue = Annotated[
+    str,
+    BeforeValidator(lambda v: "".join(v) if isinstance(v, list) else v),
+]
+
+
+class LLMProvider(BaseModel):
+    """Model representing a provider of Large Language Models (LLMs)."""
+
+    hrid: str
+    base_url: SettingEnvValue
+    api_key: SettingEnvValue
+    kind: Literal["openai", "mistral"] = "openai"
+
+
+class LLMProfile(BaseModel):
+    """Based on pydantic_ai.profiles.ModelProfile to allow customization."""
+
+    supports_tools: bool | None = None
+    supports_json_schema_output: bool | None = None
+    supports_json_object_output: bool | None = None
+    default_structured_output_mode: str | None = None
+    prompted_output_template: str | None = None
+    json_schema_transformer: ImportString | None = Field(default=None, validate_default=True)
+    thinking_tags: tuple[str, str] | None = None
+    ignore_streamed_leading_whitespace: bool | None = None
+
+    # openai specific settings: should find a way to auto declare these
+    # based on OpenAIModelProfile.
+    openai_supports_strict_tool_definition: bool | None = None
+    openai_unsupported_model_settings: Sequence[str] | None = None
+    openai_supports_tool_choice_required: bool | None = None
+    openai_system_prompt_role: str | None = None
+    openai_chat_supports_web_search: bool | None = None
+    openai_supports_encrypted_reasoning_content: bool | None = None
+
+    @field_validator("json_schema_transformer", mode="after")
+    @classmethod
+    def validate_json_schema_transformer(
+        cls, value: JsonSchemaTransformer | None
+    ) -> Optional[JsonSchemaTransformer]:
+        """Convert the tools if it's a setting or environment variable."""
+        if not value:
+            return None
+
+        if issubclass(value, JsonSchemaTransformer):
+            return value
+
+        raise ValueError(f"Invalid JSON Schema Transformer '{value}'")
+
+
+class LLMSettings(BaseModel):
+    """Based on pydantic_ai.settings.ModelSettings to allow customization."""
+
+    max_tokens: int | None = None
+    temperature: float | None = None
+    top_p: float | None = None
+    timeout: float | None = None
+    parallel_tool_calls: bool | None = None
+    seed: int | None = None
+    presence_penalty: float | None = None
+    frequency_penalty: float | None = None
+    logit_bias: dict[str, int] | None = None
+    stop_sequences: list[str] | None = None
+    extra_headers: dict[str, str] | None = None
+    extra_body: dict[str, str] | None = None
+
+
+class LLModel(BaseModel):
+    """Model representing a Large Language Model (LLM)."""
+
+    hrid: str
+    model_name: SettingEnvValue
+    human_readable_name: str
+    profile: LLMProfile | None = None
+    provider_name: str | None = None
+    provider: LLMProvider | None = None
+    settings: LLMSettings | None = None
+    is_active: bool
+    icon: LongStringAsListValue | None = None
+    supports_streaming: bool | None = None
+    system_prompt: SettingEnvValue
+    tools: list[str]
+
+    @field_validator("tools", mode="before")
+    @classmethod
+    def validate_tools(cls, value: list[str] | str) -> list[str]:
+        """Convert the tools if it's a setting or environment variable."""
+        if isinstance(value, str):
+            return _get_setting_or_env_or_value(value)
+        return value
+
+    @model_validator(mode="after")
+    def check_provider_or_provider_name(self) -> Self:
+        """
+        Do some validation regarding provider and provider_name:
+        - Either `provider_name` or `provider` must be set, but not both.
+        - If neither is set, `model_name` must be in the format '<provider>:<model>'.
+        """
+        if bool(self.provider_name) and bool(self.provider):
+            raise ValueError("Either 'provider_name' or 'provider' must be set, but not both.")
+        if not self.provider_name and not self.provider and len(self.model_name.split(":")) != 2:
+            raise ValueError(
+                "Either 'provider_name' or 'provider' must be set, "
+                "unless model_name starts with '<provider>:'."
+            )
+        return self
+
+    @property
+    def is_custom(self) -> bool:
+        """Return True if the model is a custom model (i.e., defines a provider)."""
+        return self.provider is not None
+
+
+class LLMConfiguration(BaseModel):
+    """Model representing the entire LLM configuration."""
+
+    models: list[LLModel]
+    providers: list[LLMProvider]
+
+    @model_validator(mode="after")
+    def fill_providers(self) -> Self:
+        """Fill in the `provider` field of each model based on `provider_name`."""
+        provider_map = {provider.hrid: provider for provider in self.providers}
+        for model in self.models:
+            if model.provider_name:
+                try:
+                    model.provider = provider_map[model.provider_name]
+                except KeyError as exc:
+                    raise ValueError(
+                        f"Provider '{model.provider_name}' not found "
+                        f"for model '{model.model_name}'."
+                    ) from exc
+        return self
+
+
+def _read_llm_configuration(llm_configuration_file_path) -> LLMConfiguration:
+    """Read the LLM configuration from a JSON file and return an LLMConfiguration instance."""
+    with open(llm_configuration_file_path, "rb") as f:
+        data = f.read().decode("utf-8")
+    return LLMConfiguration.model_validate_json(data)
+
+
+def load_llm_configuration(llm_configuration_file_path) -> dict[str, LLModel]:
+    """Load the LLM configuration and return a mapping of model HRIDs to LLModel instances."""
+    configuration = _read_llm_configuration(llm_configuration_file_path)
+    model_map = {model.hrid: model for model in configuration.models}
+    return model_map
+
+
+@lru_cache(maxsize=1)
+def cached_load_llm_configuration(llm_configuration_file_path) -> dict[str, LLModel]:
+    """Load the LLM configuration with caching to avoid redundant loading."""
+    return load_llm_configuration(llm_configuration_file_path)

src/backend/chat/malware_detection.py

@@ -0,0 +1,52 @@
+"""Malware detection callbacks"""
+
+import logging
+
+from core.file_upload.enums import AttachmentStatus
+
+from chat.models import ChatConversationAttachment
+
+logger = logging.getLogger(__name__)
+security_logger = logging.getLogger("conversations.security")
+
+
+def conversation_safe_attachment_callback(file_path, *, conversation_id, **kwargs):
+    """Callback when a malware scan is completed and unsafe for a conversation attachment."""
+    logger.info("File %s for conversation %s is safe", file_path, conversation_id)
+
+    ChatConversationAttachment.objects.filter(
+        conversation_id=conversation_id, key=file_path
+    ).update(upload_state=AttachmentStatus.READY)
+
+
+def unknown_attachment_callback(file_path, error_info, *, conversation_id, **kwargs) -> bool:
+    """Callback when a malware scan is completed and unknown for a conversation attachment."""
+    security_logger.warning(
+        "File %s for conversation %s has an unknown reportstatus. Error info: %s",
+        file_path,
+        conversation_id,
+        error_info,
+    )
+
+    error_code = error_info.get("error_code")
+    if error_code == 413:
+        ChatConversationAttachment.objects.filter(
+            conversation_id=conversation_id, key=file_path
+        ).update(upload_state=AttachmentStatus.FILE_TOO_LARGE_TO_ANALYZE)
+        return True
+
+    return False
+
+
+def conversation_unsafe_attachment_callback(file_path, error_info, *, conversation_id, **kwargs):
+    """Callback when a malware scan is completed and unsafe for a conversation attachment."""
+    security_logger.warning(
+        "File %s for conversation %s is infected with malware. Error info: %s",
+        file_path,
+        conversation_id,
+        error_info,
+    )
+
+    ChatConversationAttachment.objects.filter(
+        conversation_id=conversation_id, key=file_path
+    ).update(upload_state=AttachmentStatus.SUSPICIOUS)

src/backend/chat/mcp_servers.py

@@ -1,6 +1,6 @@
 """MCP servers configuration: will be replaced by models."""
 
-from agents.mcp import MCPServerStreamableHttp, MCPServerStreamableHttpParams
+from pydantic_ai.mcp import MCPServerStreamableHTTP
 
 MCP_SERVERS = {
     "mcpServers": {
@@ -15,9 +15,6 @@ MCP_SERVERS = {
 def get_mcp_servers():
     """Retrieve MCP servers configuration."""
     return [
-        MCPServerStreamableHttp(
-            name=name,
-            params=MCPServerStreamableHttpParams(**server),
-        )
-        for name, server in MCP_SERVERS["mcpServers"].items()
+        MCPServerStreamableHTTP(**server_config)
+        for _name, server_config in MCP_SERVERS["mcpServers"].items()
     ]

src/backend/chat/migrations/0001_initial.py

@@ -1,14 +1,24 @@
-# Generated by Django 5.2.3 on 2025-06-26 12:15
+# Generated by Django 5.2.3 on 2025-08-06 16:42
 
 import uuid
 
+import django.core.serializers.json
+import django.db.models.deletion
+from django.conf import settings
 from django.db import migrations, models
 
+import django_pydantic_field.compat.django
+import django_pydantic_field.fields
+
+import chat.ai_sdk_types
+
 
 class Migration(migrations.Migration):
     initial = True
 
-    dependencies = []
+    dependencies = [
+        migrations.swappable_dependency(settings.AUTH_USER_MODEL),
+    ]
 
     operations = [
         migrations.CreateModel(
@@ -59,19 +69,24 @@ class Migration(migrations.Migration):
                     ),
                 ),
                 (
-                    "openai_messages",
+                    "pydantic_messages",
                     models.JSONField(
                         blank=True,
                         default=list,
-                        help_text="OpenAI messages for the chat conversation, not used",
+                        help_text="Pydantic messages for the chat conversation, used for history",
                     ),
                 ),
                 (
                     "messages",
-                    models.JSONField(
+                    django_pydantic_field.fields.PydanticSchemaField(
                         blank=True,
+                        config=None,
                         default=list,
+                        encoder=django.core.serializers.json.DjangoJSONEncoder,
                         help_text="Stored messages for the chat conversation, sent to frontend",
+                        schema=django_pydantic_field.compat.django.GenericContainer(
+                            list, (chat.ai_sdk_types.UIMessage,)
+                        ),
                     ),
                 ),
                 (
@@ -82,6 +97,22 @@ class Migration(migrations.Migration):
                         help_text="Agent usage for the chat conversation, provided by OpenAI API",
                     ),
                 ),
+                (
+                    "collection_id",
+                    models.CharField(
+                        blank=True,
+                        help_text="Collection ID for the conversation, used for RAG document search",
+                        null=True,
+                    ),
+                ),
+                (
+                    "owner",
+                    models.ForeignKey(
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="conversations",
+                        to=settings.AUTH_USER_MODEL,
+                    ),
+                ),
             ],
             options={
                 "abstract": False,

src/backend/chat/migrations/0002_fix_source_type_in_messages.py

@@ -0,0 +1,54 @@
+# Generated by Django 5.2.3 on 2025-09-15 11:47
+from django.db import migrations
+
+
+def forwards_func(apps, schema_editor):
+    """Use raw SQL to replace "source_type" with "sourceType" in the messages field."""
+    with schema_editor.connection.cursor() as cursor:
+        cursor.execute("""
+            UPDATE chat_chatconversation 
+            SET messages = REPLACE(messages::text, '"source_type"', '"sourceType"')::jsonb
+            WHERE messages::text LIKE '%"source_type"%'
+        """)
+    with schema_editor.connection.cursor() as cursor:
+        cursor.execute("""
+            UPDATE chat_chatconversation
+            SET ui_messages = REPLACE(ui_messages::text, '"source_type"', '"sourceType"')::jsonb
+            WHERE ui_messages::text LIKE '%"source_type"%'
+        """)
+
+
+def reverse_func(apps, schema_editor):
+    """Use raw SQL to replace "sourceType" with "source_type" in the messages field."""
+    with schema_editor.connection.cursor() as cursor:
+        cursor.execute("""
+            UPDATE chat_chatconversation 
+            SET messages = REPLACE(messages::text, '"sourceType"', '"source_type"')::jsonb
+            WHERE messages::text LIKE '%"sourceType"%'
+        """)
+    with schema_editor.connection.cursor() as cursor:
+        cursor.execute("""
+            UPDATE chat_chatconversation
+            SET ui_messages = REPLACE(ui_messages::text, '"sourceType"', '"source_type"')::jsonb
+            WHERE ui_messages::text LIKE '%"sourceType"%'
+        """)
+
+
+class Migration(migrations.Migration):
+    """
+    Rename source_type to sourceType in messages field of ChatConversation model.
+
+    Warning: This migration is not fail-safe, if the messages field contains
+    other occurrences of "source_type" or "sourceType" in other contexts, they will
+    also be replaced. Also, if the messages field is very large, this migration
+    may take a long time to run.
+    => OK because we are only in development phase.
+    """
+
+    dependencies = [
+        ("chat", "0001_initial"),
+    ]
+
+    operations = [
+        migrations.RunPython(forwards_func, reverse_func),
+    ]

src/backend/chat/migrations/0002_initial.py

@@ -1,26 +0,0 @@
-# Generated by Django 5.2.3 on 2025-06-26 12:15
-
-import django.db.models.deletion
-from django.conf import settings
-from django.db import migrations, models
-
-
-class Migration(migrations.Migration):
-    initial = True
-
-    dependencies = [
-        ("chat", "0001_initial"),
-        migrations.swappable_dependency(settings.AUTH_USER_MODEL),
-    ]
-
-    operations = [
-        migrations.AddField(
-            model_name="chatconversation",
-            name="owner",
-            field=models.ForeignKey(
-                on_delete=django.db.models.deletion.CASCADE,
-                related_name="conversations",
-                to=settings.AUTH_USER_MODEL,
-            ),
-        ),
-    ]

src/backend/chat/migrations/0003_chatconversationcontext.py

@@ -0,0 +1,78 @@
+# Generated by Django 5.2.3 on 2025-09-17 12:58
+
+import uuid
+
+import django.db.models.deletion
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+    dependencies = [
+        ("chat", "0002_fix_source_type_in_messages"),
+    ]
+
+    operations = [
+        migrations.CreateModel(
+            name="ChatConversationContext",
+            fields=[
+                (
+                    "id",
+                    models.UUIDField(
+                        default=uuid.uuid4,
+                        editable=False,
+                        help_text="primary key for the record as UUID",
+                        primary_key=True,
+                        serialize=False,
+                        verbose_name="id",
+                    ),
+                ),
+                (
+                    "created_at",
+                    models.DateTimeField(
+                        auto_now_add=True,
+                        help_text="date and time at which a record was created",
+                        verbose_name="created on",
+                    ),
+                ),
+                (
+                    "updated_at",
+                    models.DateTimeField(
+                        auto_now=True,
+                        help_text="date and time at which a record was last updated",
+                        verbose_name="updated on",
+                    ),
+                ),
+                (
+                    "kind",
+                    models.CharField(
+                        choices=[("image", "Image"), ("document", "Document")],
+                        help_text="Kind of the chat conversation context (e.g., 'image', 'document')",
+                        max_length=50,
+                    ),
+                ),
+                (
+                    "name",
+                    models.CharField(
+                        help_text="Key of the chat conversation context", max_length=100
+                    ),
+                ),
+                (
+                    "content",
+                    models.TextField(
+                        blank=True, help_text="Value of the chat conversation context", null=True
+                    ),
+                ),
+                (
+                    "conversation",
+                    models.ForeignKey(
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="contexts",
+                        to="chat.chatconversation",
+                    ),
+                ),
+            ],
+            options={
+                "unique_together": {("conversation", "name")},
+            },
+        ),
+    ]

src/backend/chat/migrations/0004_chatconversationattachment_and_more.py

@@ -0,0 +1,100 @@
+# Generated by Django 5.2.7 on 2025-10-17 16:10
+
+import uuid
+
+import django.db.models.deletion
+from django.conf import settings
+from django.db import migrations, models
+
+import core.file_upload.enums
+
+
+class Migration(migrations.Migration):
+    dependencies = [
+        ("chat", "0003_chatconversationcontext"),
+        migrations.swappable_dependency(settings.AUTH_USER_MODEL),
+    ]
+
+    operations = [
+        migrations.CreateModel(
+            name="ChatConversationAttachment",
+            fields=[
+                (
+                    "id",
+                    models.UUIDField(
+                        default=uuid.uuid4,
+                        editable=False,
+                        help_text="primary key for the record as UUID",
+                        primary_key=True,
+                        serialize=False,
+                        verbose_name="id",
+                    ),
+                ),
+                (
+                    "created_at",
+                    models.DateTimeField(
+                        auto_now_add=True,
+                        help_text="date and time at which a record was created",
+                        verbose_name="created on",
+                    ),
+                ),
+                (
+                    "updated_at",
+                    models.DateTimeField(
+                        auto_now=True,
+                        help_text="date and time at which a record was last updated",
+                        verbose_name="updated on",
+                    ),
+                ),
+                (
+                    "upload_state",
+                    models.CharField(
+                        choices=core.file_upload.enums.AttachmentStatus.choices,
+                        default=core.file_upload.enums.AttachmentStatus["PENDING"],
+                        max_length=40,
+                    ),
+                ),
+                (
+                    "key",
+                    models.CharField(help_text="File path of the attachment in the object storage"),
+                ),
+                ("file_name", models.CharField(help_text="Original name of the attachment file")),
+                (
+                    "content_type",
+                    models.CharField(help_text="MIME type of the attachment file", max_length=100),
+                ),
+                ("size", models.PositiveBigIntegerField(blank=True, null=True)),
+                (
+                    "conversation",
+                    models.ForeignKey(
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="attachments",
+                        to="chat.chatconversation",
+                    ),
+                ),
+                (
+                    "uploaded_by",
+                    models.ForeignKey(
+                        help_text="User who uploaded the attachment",
+                        on_delete=django.db.models.deletion.PROTECT,
+                        related_name="uploaded_attachments",
+                        to=settings.AUTH_USER_MODEL,
+                    ),
+                ),
+                (
+                    "conversion_from",
+                    models.CharField(
+                        blank=True,
+                        help_text="Original file key if the Markdown from another file",
+                        null=True,
+                    ),
+                ),
+            ],
+            options={
+                "abstract": False,
+            },
+        ),
+        migrations.DeleteModel(
+            name="ChatConversationContext",
+        ),
+    ]