Configuration reference

Compose provides working local defaults. Copy .env.example to .env only when you need custom ports, models, parsers, reliability settings, or production storage.

Network and runtime mode

Variable	Default	Description
WEB_PORT	3000	Web port on the host
API_PORT	8000	API port on the host
BIND_ADDRESS	127.0.0.1	Listen address; configure perimeter protection before using 0.0.0.0 on a server
SAG_ENVIRONMENT	dev	Use prod in production
SAG_CORS_ORIGINS	http://localhost:3000	Frontend origins allowed to call the API with credentials
NEXT_PUBLIC_API_BASE	http://localhost:8000	Build-time API address used by the browser
SAG_SECRET_KEY	Local development default	JWT signing secret; replace in production
SAG_ALLOW_REGISTRATION	false	Whether the traditional email registration endpoint is open

LLM

Variable	Default	Description
SAG_LLM_PROVIDER	openai	LiteLLM provider identifier
SAG_LLM_BASE_URL	https://api.302ai.cn/v1	OpenAI-compatible endpoint; native providers may leave it empty
SAG_LLM_API_KEY	Empty	Model key
SAG_LLM_MODEL	qwen3.6-flash	Server-side model identifier
SAG_LLM_MAX_TOKENS	20000	Maximum generated tokens
SAG_LLM_TIMEOUT_MS	60000	Per-request timeout in milliseconds
SAG_LLM_MAX_RETRIES	2	Number of model-request retries

The UI and API start with an empty key, but event extraction, query understanding, precise-retrieval reranking, and answer generation remain unavailable.

Embedding

Variable	Default	Description
SAG_EMBEDDING_MODEL	bge-large-en-v1.5	Embedding model identifier
SAG_EMBEDDING_BASE_URL	https://api.302ai.cn/v1	Separate Embedding endpoint
SAG_EMBEDDING_API_KEY	Empty	Separate key; application settings can reuse the LLM configuration

Embedding is required for document vectorization and fast retrieval. Existing vectors do not move automatically into the semantic space of a new model. Reprocess documents after changing models.

Parsing and chunking

Variable	Default	Description
SAG_DOCUMENT_PARSER	auto	Select MinerU or MarkItDown automatically
SAG_MINERU_BASE_URL	https://api.302ai.cn	MinerU service URL
SAG_MINERU_API_KEY	Empty	MinerU key
SAG_MINERU_VERSION	2.5	MinerU version
SAG_DOCUMENT_EXTRACT_CONCURRENCY	5	Parallel extraction count
SAG_DOCUMENT_CHUNK_MAX_TOKENS	1000	Target upper bound for one chunk
SAG_DOCUMENT_CHUNK_MODE	standard	Chunking mode

Background jobs and engine

Variable	Default	Description
SAG_JOB_MAX_ATTEMPTS	3	Maximum attempts for a background job
SAG_ENGINE_CACHE_SIZE	16	Engine-instance cache capacity
SAG_ENGINE_WARMUP_COUNT	4	Number of instances warmed on startup
SAG_MINERU_POLL_TIMEOUT	300	Optional MinerU polling timeout

PostgreSQL

Variable	Default	Description
POSTGRES_USER	sag	Database user
POSTGRES_PASSWORD	None	Required production override
POSTGRES_DB	sag	Database name

The production override also sets SAG_SAG_RELATIONAL_PROVIDER=postgres and SAG_SAG_VECTOR_PROVIDER=pgvector. Do not create a partial storage combination by overriding only one half in the default Compose setup.