Codex Local
c5f921243e
- LLM_API_KEY support across all LLM calls - TEI embedding server (bge-m3, :9500) - Hierarchical AGENTS.md documentation - CLAUDE.md updated for new architecture
59 lines
3.4 KiB
Bash
59 lines
3.4 KiB
Bash
# ---------------------------------------------------------------------------
|
|
# KXKM_Clown — Environment Variables
|
|
# ---------------------------------------------------------------------------
|
|
# Copy this file to .env and adjust values for your deployment.
|
|
# cp .env.example .env
|
|
# ---------------------------------------------------------------------------
|
|
|
|
# --- Local LLM runtime (vLLM / TurboQuant) ---------------------------------
|
|
# Primary chat/completion runtime. Must expose an OpenAI-compatible API.
|
|
LLM_URL=http://host.docker.internal:11434
|
|
LLM_MODEL=qwen-14b-awq
|
|
LLM_API_KEY= # Bearer token for vLLM --api-key (leave empty for Ollama)
|
|
|
|
# --- Embeddings backend (auxiliary) ----------------------------------------
|
|
# TEI (recommended): dedicated embedding server on port 9500
|
|
# Ollama: fallback, uses /api/embed on port 11434
|
|
OLLAMA_URL=http://host.docker.internal:9500
|
|
EMBEDDING_BACKEND=tei # "tei" (OpenAI /v1/embeddings) or "ollama" (/api/embed)
|
|
RAG_EMBEDDING_MODEL=BAAI/bge-m3 # Model name (must match TEI --model-id or Ollama model)
|
|
|
|
# --- Ports ------------------------------------------------------------------
|
|
APP_PORT=3333 # V1 Express server
|
|
API_PORT=4180 # V2 API server
|
|
PG_PORT=5432 # PostgreSQL (exposed to host)
|
|
# LLM_PORT=11434 # Only needed if you expose the local runtime from compose
|
|
|
|
# --- Admin ------------------------------------------------------------------
|
|
ADMIN_BOOTSTRAP_TOKEN= # Initial admin token (set a strong secret)
|
|
ADMIN_ALLOWED_SUBNETS= # CIDR subnets for admin access (e.g. 192.168.1.0/24)
|
|
OWNER_NICK= # Owner nickname in chat
|
|
|
|
# --- Chat -------------------------------------------------------------------
|
|
MAX_GENERAL_RESPONDERS=4 # Max personas responding in #general
|
|
|
|
# --- Vision (image analysis in chat) ----------------------------------------
|
|
# VISION_MODEL=qwen3-vl:8b # Vision model used by the configured runtime
|
|
|
|
# --- Training (Node Engine worker) ------------------------------------------
|
|
# PYTHON_BIN=/home/kxkm/venv/bin/python3 # Python with ML libs (PyTorch, Unsloth, TRL)
|
|
# TRAINING_TIMEOUT_MS=3600000 # Training timeout (default 1h)
|
|
# SCRIPTS_DIR=/app/scripts # Path to train_unsloth.py, eval_model.py
|
|
# Note: GPU passthrough is enabled by default in docker-compose (deploy.resources)
|
|
# Note: Host venv is mounted read-only at /home/kxkm/venv in the worker container
|
|
|
|
# --- ComfyUI (image generation) --------------------------------------------
|
|
# COMFYUI_URL=http://localhost:8188 # ComfyUI API endpoint for /imagine command
|
|
|
|
# --- Mascarade (optional cloud/provider routing) ----------------------------
|
|
# MASCARADE_URL=http://127.0.0.1:8100 # Optional cloud/provider router
|
|
# MASCARADE_API_KEY= # API key for authenticated endpoints (/agents, /orchestrate)
|
|
|
|
# --- External services (optional) ------------------------------------------
|
|
# WEB_SEARCH_API_BASE= # Web search API endpoint for /web command
|
|
# SEARXNG_URL=http://localhost:8080 # SearXNG instance for web search (fallback: DuckDuckGo)
|
|
|
|
# --- Discord Bot (optional, --profile discord) --------------------------------
|
|
# DISCORD_BOT_TOKEN= # Discord bot token (from Discord Developer Portal)
|
|
# DISCORD_CHANNEL_ID= # Discord channel ID to bridge with KXKM chat
|