.env.example

# LLM (build / answer / judge)
SGAO_API_KEY=your-api-key-here
SGAO_BASE_URL=https://sub2api.sgao.me/v1
# Verified model options on the SGAO OpenAI-compatible endpoint.
SGAO_MODEL=gpt-5.4-mini
SGAO_MODEL_GPT54_MINI=gpt-5.4-mini
SGAO_MODEL_GPT56_LUNA=gpt-5.6-luna
LLM_REQUEST_PROFILE=openai

# GraphMem V2 fixed local embedding service
EMBEDDING_BASE_URL=http://127.0.0.1:8001/v1
EMBEDDING_MODEL=Qwen3-Embedding-0.6B
EMBEDDING_API_KEY=replace-with-local-embedding-key

# Optional: OpenAI-compatible custom endpoint (overrides DEEPSEEK_* when set)
# BASE_URL=https://your-endpoint/v1
# MODEL=gpt-4o-mini
# DEEPSEEK_API_KEY=your-api-key-here

# Legacy local vLLM scripts may use their own LLM_PORT / EMBED_PORT values.
# LLM_PORT=8001
# EMBED_PORT=8002

# HuggingFace cache for LLMLingua compression model downloads
# HF_HOME=~/.cache/huggingface
# RUN_HF_HOME=~/.cache/huggingface