API reference - Belvedir

The inference endpoints run your LLM calls: point an OpenAI-compatible or Anthropic client at Belvedir with your Belvedir key and each call is metered and billed per token. The tracing endpoints receive OpenTelemetry spans and ground-truth outcomes; the SDK calls those for you, and any stock OpenTelemetry exporter can target them too.

Base URL: https://platform.belvedir.ai
Authentication: API key via Authorization: Bearer or x-api-key header. See Authentication.

Inference

Endpoint	What it does
`POST /api/v1/route/chat/completions`	OpenAI-compatible chat completions, routed or anchored on the model your code names
`POST /api/v1/messages`	The Anthropic Messages API, passed through unchanged
`POST /api/v1/route/embeddings`	OpenAI-compatible embeddings
`/api/v1/route/batches`	Batch inference: create, list, status, results, cancel, and `/uploads` for large submissions; the Anthropic SDK facade at `/api/v1/messages/batches` is on the same page
`POST /api/v1/route`	A routing decision without execution, for calls your own code runs

To get started, point your existing client at Belvedir: Route your calls.

Tracing

Endpoint	What it does
`POST /api/v1/traces`	Ingest OpenTelemetry spans as OTLP over HTTP (JSON or protobuf)
`POST /api/v1/outcomes`	Report whether a session actually worked, from a signal only your app has