.env.example

# Copy to .env before `docker compose up` (optional).

API_PORT=5100
ADMIN_PORT=5200
PUBLIC_API_URL=http://localhost:5100
ADMIN_PUBLIC_URL=http://localhost:5200

# --- LLM engine -------------------------------------------------------------
# Bundled engines (no host LLM needed):
#   llama.cpp: docker compose -f docker-compose.yml -f docker-compose.llamacpp.yml up --build
#   vLLM:      docker compose -f docker-compose.yml -f docker-compose.vllm.yml up --build
# The overrides set LLM_ENDPOINT for you; the values below only matter for the plain compose file.

# Model name sent to the engine. llama-server ignores it; vLLM uses it as --served-model-name;
# Ollama needs a pulled tag (e.g. qwen3.5:9b). Overrides default to local-model, plain compose to qwen3.5:9b.
# DEFAULT_LLM_MODEL=local-model

# llama.cpp override
# LLAMACPP_HF_MODEL=bartowski/Qwen2.5-3B-Instruct-GGUF:Q4_K_M
# LLAMACPP_CTX_SIZE=8192
# LLAMACPP_IMAGE=ghcr.io/ggml-org/llama.cpp:server

# vLLM override
# VLLM_MODEL=Qwen/Qwen2.5-7B-Instruct
# VLLM_MAX_MODEL_LEN=16384
# VLLM_TOOL_CALL_PARSER=hermes
# HF_TOKEN=

# Engine already running on the host (any OpenAI-compatible /v1 server):
# LLM_ENDPOINT=http://host.docker.internal:8080

# Ollama on the host (used only when LLM_ENDPOINT is empty):
OLLAMA_ENDPOINT=http://host.docker.internal:11434

# Host defaults for apps with llmBackend openai / openai-compatible / vllm and no per-app endpoint
# OPENAI_ENDPOINT=https://api.openai.com
# OPENAI_API_KEY=

# --- Keys (change before any real use) -------------------------------------
MASTER_KEY=cm_master_dev_key_change_me
DEMO_APP_API_KEY=cm_live_dev_key_change_me

# File (default) or Postgres — for Postgres set ConnectionStrings__ContextMemory in compose override
PERSISTENCE_PROVIDER=File