Skip to content

API Server

Set up and configure Backend.AI GO's OpenAI-compatible API server for integrating local LLMs with external applications.

The Management API (default http://127.0.0.1:8001/api/v1) sits beside it and controls the app itself. Its full endpoint list is browsable in Swagger UI at /api/docs, and the pages below cover the parts that need more than a schema:

  • Event Streams: the SSE and WebSocket transports over the event bus, resuming by event id, and how gaps and lag are reported.
  • Continuum Router: the request-routing layer in front of local and cloud models.
  • External Access: binding the server to a non-loopback address safely.
  • Model Hub Mirror: serving model downloads from a mirror.
  • Settings → Claude Code: model aliases, web search, and the verify probe.
  • MCP Endpoint: exposing Backend.AI GO's built-in tools to external MCP clients.
  • Autonomous Agents: the 34 provider and Hermes endpoints, runtime availability, events, scopes, and operational walkthroughs.
  • Automations: the /schedules and /executions endpoints behind the Automations page, including export, import, and the credential warning that goes with them.
  • Squads: the 95-endpoint multi-agent squad surface, its events, and its walkthroughs.
  • Data Hub: the 66-endpoint document corpus surface, its data:* events, and the request fields that look like they exist but do not.
  • Memory: the 27-endpoint persistent fact store, its memory:* events, whole-store export and import, and the raised body limit that import alone carries.
  • Router Statistics Coverage: what the statistics page counts and what it does not.