Stable client API
OpenAI Chat, OpenAI Responses, Anthropic Messages, SDKs, Codex CLI, and Claude Code can call one governed endpoint.
Metrum AI Product Documentation
Metrum AI Router
An open-source LLM smart router that selects a different upstream model per request from a deployment-owned policy, then records why. Gateway functions underneath routing keep GenAI clients stable while provider keys, budgets, telemetry, and request-time cost accounting stay under platform control.
Policy-owned model groups
Expose quality and cost contracts instead of raw provider IDs, then tune providers and weights without client rewrites.
Request-shape safety
Tool, image, dialect, and max-token requirements filter targets before the router calls an upstream model.
Cost and performance evidence
Usage rows and reports preserve provider/model, tokens, image fields, cache, latency, attempts, fallbacks, and request-time cost.
Private and hosted upstreams
Mix external providers with enterprise vLLM, SGLang, Baseten-style, or other OpenAI-compatible services behind one API.