If you write the code
“How do I send one request and see why it went where it went?”pip or npm install, or just point your OpenAI client at a new base URL. Every answer carries the target, the confidence, the alternatives, the cost and a trace id; report the outcome and the next decision moves. Claude, Cursor and VS Code can call the same router over MCP.
How the promise is kept
Every line below names the page or endpoint that does the work. Open it before you believe it.
Python SDK with zero dependencies, TypeScript SDK typed to every endpoint, REST and OpenAI-compatible clients
SDKsType a request in the playground and watch signals, admissible targets, the ranking and the cost
PlaygroundA trace per request - what it saw, who was allowed to answer, how they ranked, why this one - and OpenTelemetry export
ObservabilityThe MCP server: route, estimate, explain, ask and feedback from any MCP client with your workspace key
MCPAgents, skills, personas, tools, prompts and templates from the marketplace join the routing pool like a model
MarketplaceWhat you leave with
A decision object: target, confidence, signals, policy rejections, ranked candidates with reasons, and the trace to replay it.
The other readers
The free plan is enough to import a bill, route a first request and read its trace. When the numbers hold, the plans and the agreement follow.
Free plan, no card. Fifteen thousand decisions a month with the full trace.