LLM Providers#

Use these pages to choose which provider and which model each OSPREY model call uses. You make two separate decisions:

  • Which provider. A provider is an endpoint and its API key. It can be an Anthropic-native endpoint (Anthropic itself or an institutional proxy), an OpenAI-compatible endpoint, or a model served on your own hardware. You declare providers in one catalog, providers.yml beside the deployment’s profile.yml, which is rendered into api.providers.

  • Which model handles which task. This is routing. OSPREY routes for two consumers, and you choose the provider and model for each one separately:

    • The OSPREY agent speaks the Anthropic Messages API. Set its provider and model in the profile’s top-level provider: and model: fields.

    • The MCP (Model Context Protocol) tool servers call the same named providers directly through LiteLLM. Set each tool server’s own provider setting. Some of these default to the agent’s.

Both consumers draw endpoints and keys from the same catalog, so a key you set there serves both.

Configure LLM Providers

The available providers, the API key each one expects, the providers.yml catalog that declares them, and the profile fields that point the agent at one.

Configure LLM Providers
Run Open & Local Models

Run the agent on open-weight or self-hosted models. Covers how the local translation proxy bridges an OpenAI-protocol endpoint to the Anthropic Messages API, remote or on your own machine. Also lists the open models known to sustain the full end-to-end (e2e) test suite, and how to benchmark one yourself.

Run Open & Local Models