Skip to main content
Astrolabe is a self-hosted AI gateway for OpenClaw. It keeps routing, fallback behavior, model cost controls, and safety policy in one place while staying simple to deploy:
  • no database
  • no hosted control plane
  • no runtime model discovery
  • one OpenRouter key
  • one Astrolabe API key

​
Core ideas

  1. POST /v1/responses is the primary interface.
  2. POST /v1/chat/completions remains as a compatibility layer.
  3. Virtual models are the user-facing surface.
  4. Static manifests control the raw roster and lane candidates.
  5. m27 is the main workhorse.
  6. m25 is strict-budget and fallback only.

​
What Astrolabe owns

  • request classification
  • lane routing
  • specialist promotion
  • retry and fallback behavior
  • non-stream verification
  • tool-policy checks
  • untrusted-content safety floors
  • route transparency through headers and metadata

​
Default virtual models

  • astrolabe/auto
  • astrolabe/coding
  • astrolabe/research
  • astrolabe/vision
  • astrolabe/strict-json
  • astrolabe/cheap
  • astrolabe/safe

Quickstart

Run Astrolabe locally and verify Responses-first routing.

Connect OpenClaw

Point OpenClaw at Astrolabe as its default provider.

Configuration

Tune profiles, safety controls, and Responses URL policy.

Routing Reference

See the current lane logic, workhorse model, and guardrails.

Model Roster

Review the static roster and alias map.

API Reference

Check endpoint contracts for /v1/responses, /v1/chat/completions, and metadata endpoints.