Skip to main content

Settings and Authentication

Settings​

Module: atlasml/config.py. Settings are strongly typed via Pydantic and loaded from environment variables. The service can operate with safe defaults in tests or when explicitly requested, while production requires explicit configuration.

Environment variables:

  • ATLAS_API_KEYS: Comma-separated list of API key tokens (e.g., key1,key2)
  • WEAVIATE_HOST: Full HTTPS URL of centralized Weaviate instance (e.g., https://weaviate.example.com)
  • WEAVIATE_PORT: Weaviate HTTPS port (always 443 for centralized setup)
  • WEAVIATE_API_KEY: API key for authenticating with centralized Weaviate (REST API only)
  • OPENAI_API_KEY: Azure OpenAI API key for generating embeddings
  • OPENAI_API_URL: Azure OpenAI endpoint URL
  • SENTRY_DSN: Optional Sentry DSN (used when ENV=production)
  • ENV: Environment name (e.g., dev, production)
Centralized Weaviate Required

AtlasML requires the centralized Weaviate setup with HTTPS and API key authentication. See the Admin Configuration Guide for complete setup instructions.

Defaults are used in tests or when get_settings(use_defaults=True) is explicitly requested. The ENV variable also toggles optional production-only features like Sentry error reporting.

Authentication​

Module: atlasml/dependencies.py. Authentication uses a simple API-key strategy implemented as FastAPI dependencies. This can be attached per-route or globally depending on deployment needs.

AtlasML uses a simple API key mechanism via the Authorization header. Keys are compared verbatim and should be rotated and stored securely via your deployment platform’s secret manager.

  • Header: Authorization: <API_KEY>
  • Keys are read from ATLAS_API_KEYS and compared verbatim.
  • Use Depends(TokenValidator) or validate_token on routes to enforce auth. For example, add dependencies=[Depends(TokenValidator)] to a router or endpoint.