Skip to main content
Most Astrolabe behavior is configured directly in server.js.

​
Model tiers

const tierModels = {
  TIER0: "openai/gpt-5-nano",
  TIER1: "x-ai/grok-4.1-fast",
  TIER2: "anthropic/claude-opus-4.6"
};
  • TIER0: classifier + self-check model
  • TIER1: default serving tier
  • TIER2: premium fallback tier

​
Cost estimation table

const pricePer1M = {
  "openai/gpt-5-nano": { input: 0.05, output: 0.4 },
  "x-ai/grok-4.1-fast": { input: 0.2, output: 0.5 },
  "anthropic/claude-opus-4.6": { input: 15, output: 75 }
};
Update this as provider pricing changes.

​
Routing and escalation

  1. classifyTier(...) selects TIER0, TIER1, or TIER2.
  2. Streaming requests are routed once and passed through.
  3. Non-streaming requests run runSelfCheck(...).
  4. If self-check confidence is low, Astrolabe retries once on TIER2.

​
Optional upstream metadata

Set these env vars if you want OpenRouter attribution headers: 
OPENROUTER_SITE_URL=https://your-site.example
OPENROUTER_APP_NAME=Astrolabe