Configuration

All settings are configured via environment variables, loaded through Pydantic BaseSettings.

Required Settings

Setting	Type	Description
`DATABASE_URL`	string	PostgreSQL connection string (e.g., `postgresql://user:pass@host:5432/dbname`)
`OPENAI_API_KEY`	string	OpenAI API key for embeddings and LLM calls

API Security

Setting	Type	Default	Description
`API_KEY`	string	`""`	REST API authentication key. If empty, API runs without auth
`CORS_ORIGINS`	list[string]	`["*"]`	Allowed CORS origins
`RATE_LIMIT_ENABLED`	boolean	`false`	Enable endpoint rate limiting

Connection Pool

Setting	Type	Default	Description
`POOL_MIN_SIZE`	integer	`2`	Minimum pool connections
`POOL_MAX_SIZE`	integer	`10`	Maximum pool connections
`POOL_TIMEOUT`	float	`5.0`	Seconds to wait for a connection
`POOL_MAX_IDLE`	float	`300.0`	Seconds before closing an idle connection
`DB_STATEMENT_TIMEOUT`	integer	`30`	Query timeout in seconds

API Server

Setting	Type	Default	Description
`API_HOST`	string	`"0.0.0.0"`	REST API bind address
`API_PORT`	integer	`8000`	REST API port
`LOG_LEVEL`	string	`"INFO"`	Logging level

Search

Setting	Type	Default	Description
`MIN_SIMILARITY_THRESHOLD`	float	`0.3`	Default minimum similarity for search results

Embeddings

Setting	Type	Default	Description
`EMBEDDING_MODEL`	string	`"text-embedding-3-small"`	OpenAI embedding model
`EMBEDDING_BATCH_SIZE`	integer	`500`	Texts per API batch call

Enrichment

Setting	Type	Default	Description
`ENRICHMENT_MODEL`	string	`"gpt-5.4-nano"`	LLM for HTS6 enrichment
`ENRICHMENT_TEMPERATURE`	float	`0.1`	LLM temperature
`ENRICHMENT_BATCH_SIZE`	integer	`50`	Codes per batch
`ENRICHMENT_MAX_WORKERS`	integer	`25`	Concurrent enrichment threads

Graph Pipeline: Edge Generation

Setting	Type	Default	Description
`GRAPH_KNN_K`	integer	`30`	Nearest neighbors per code
`GRAPH_KNN_MIN_SIMILARITY`	float	`0.65`	Minimum cosine similarity
`GRAPH_KNN_LEVEL`	string	`"hts8"`	Embedding level for KNN
`GRAPH_KNN_VARIANT`	string	`"full"`	Embedding variant for KNN

Graph Pipeline: Edge Classification

Setting	Type	Default	Description
`GRAPH_CLASSIFICATION_MODEL`	string	`"gpt-5.4-nano"`	LLM for edge classification
`GRAPH_CLASSIFICATION_TEMPERATURE`	float	`0.1`	LLM temperature
`GRAPH_CLASSIFICATION_MAX_TOKENS`	integer	`1500`	Max response tokens
`GRAPH_CLASSIFICATION_CONCURRENCY`	integer	`200`	Concurrent LLM calls
`GRAPH_CLASSIFICATION_BATCH_SIZE`	integer	`500`	Database flush batch size
`GRAPH_CLASSIFICATION_CHUNK_SIZE`	integer	`1000`	Candidates per processing chunk
`GRAPH_CLASSIFICATION_CHUNK_DELAY`	float	`1.0`	Seconds between chunks (rate limit pacing)

Example .env


# Required
DATABASE_URL=postgresql://user:password@db.supabase.co:5432/postgres
OPENAI_API_KEY=sk-...

# API Security (optional)
API_KEY=your-secret-api-key
RATE_LIMIT_ENABLED=true

# Connection Pool
POOL_MIN_SIZE=2
POOL_MAX_SIZE=10

# Search
MIN_SIMILARITY_THRESHOLD=0.3

# Graph Pipeline
GRAPH_KNN_K=30
GRAPH_CLASSIFICATION_CONCURRENCY=200