Providers & routing

llmBreakr exposes a single OpenAI-compatible surface — /v1/chat/completions — regardless of which provider actually serves the request.

Supported providers

• OpenAI
• Anthropic
• Google Gemini

Each provider is implemented as an adapter that translates the OpenAI-shaped request into that provider's native format, and translates the response (including streamed chunks) back into an OpenAI-compatible shape. Your client code never needs to know which provider is behind a given model.

More providers are planned, and the adapter pattern is intentionally built to make community-contributed providers straightforward — see Contributing.

Model resolution

Which provider handles a request is determined by the model field in the request body, resolved against what the calling project has been granted access to:

{
  "model": "gpt-4o-mini",
  "messages": [{ "role": "user", "content": "Hello" }]
}

Swap "model" for an Anthropic or Gemini model name enabled for that project, and the rest of the request is unchanged — same endpoint, same auth header, same response shape, streaming included.

Project-based access

Model access isn't global — it's scoped per project. A project might be allowed gpt-4o-mini and claude-3-5-haiku but not gpt-4o, letting you control cost exposure and capability per team or environment without maintaining separate deployments.