Configuration

What this page covers

This page focuses on environment variables you need to run and tune Postgram.

Core workflow

1. For Docker Compose, choose the embedding provider before the first start. Compose generates and persists PostgreSQL and Admin installation secrets.
2. Add a real OPENAI_API_KEY in .env only when choosing hosted OpenAI embeddings; otherwise run Ollama on the Docker host for the local path.
3. Set server/runtime overrides only when your deployment needs them.
4. Validate startup and queue behavior with health + search checks.

Server

Variable	Required	Default	Meaning
DATABASE_URL	non-Compose or external Postgres		Full Postgres connection string. Default Compose creates its own Postgres connection.
OPENAI_API_KEY	conditional		Required for OpenAI embeddings or OpenAI extraction.
PORT	no	3100	HTTP/MCP server port.
OAUTH_ENABLED	no	false	Enable OAuth authorization-code, PKCE, and Dynamic Client Registration routes for native remote MCP connectors.
PUBLIC_BASE_URL	when OAuth enabled		Public HTTPS origin for OAuth metadata and callback URLs. Example: https://postgram.example.com.
LOG_LEVEL	no	info	Logger level.
ENRICHMENT_POLL_INTERVAL_MS	no	1000	Enrichment worker polling interval.

Embeddings

Variable	Required	Default	Meaning
EMBEDDING_PROVIDER	no	Compose chooses openai with OPENAI_API_KEY, otherwise ollama	openai or ollama.
EMBEDDING_MODEL	no	provider-default	Embedding model name.
EMBEDDING_DIMENSIONS	no	provider-default	Must match active embedding model.
EMBEDDING_BASE_URL	conditionally	OLLAMA_BASE_URL fallback	For Ollama host.
EMBEDDING_API_KEY	no		Optional bearer token.

For Docker + Ollama on the host, use EMBEDDING_BASE_URL=http://host.docker.internal:11434.

LLM Extraction

Variable	Required	Default	Meaning
EXTRACTION_ENABLED	no	false	Enable extraction pass.
EXTRACTION_MEMORY_MODE	no	embed_only	embed_only keeps memories searchable without graph/entity extraction; extract_durable extracts durable memories; extract_all extracts durable memories and session context.
EXTRACTION_PROVIDER	no	openai	openai, anthropic, ollama, openai-compatible.
EXTRACTION_MODEL	no	provider-default	LLM model name for extraction.
EXTRACTION_BASE_URL	when provider-compatible		Base URL for compatible OpenAI-compatible API.
EXTRACTION_API_KEY	no		Optional bearer token for extraction API.
ANTHROPIC_API_KEY	conditionally		Required for Anthropic provider.
OLLAMA_BASE_URL	no	http://host.docker.internal:11434 in Compose	Base URL for Ollama extraction.

CLI/Admin/Backup

Variable	Required	Purpose
PGM_API_URL	yes	pgm server URL.
PGM_API_KEY	yes	pgm authentication token.
DATABASE_URL or PGM_DATABASE_URL	yes (admin)	DB access for admin commands.
PGM_BACKUP_PASSPHRASE	when encrypted backup	GPG passphrase.

Notes

• EXTRACTION_AUTO_CREATE_ENTITIES and semantic-neighbor settings allow relationship densification without re-processing everything manually.
• Memory graph extraction is opt-in. Leave EXTRACTION_MEMORY_MODE=embed_only unless your deployment intentionally wants memory-derived graph edges.
• Use admin embeddings migrate when changing vector dimensions after changing model family.
• PUBLIC_BASE_URL is the public origin only. Do not include /mcp; connector URLs use ${PUBLIC_BASE_URL}/mcp.