Model routing

Osyra's supported OpenAI-compatible surface accepts a model identifier for chat completions and embeddings. Edge carries that identifier into the Model Broker request and returns the actual model and provider reported by the completed operation.

Use the same https://api.osyra.ai/v1 base URL documented in Making requests. The verified public contract is the request and response shape on those routes—not a promise that every provider, model name, fallback sequence, or region is available in every deployment.

Provider credentials

Manage workspace provider credentials in Virtual keys. The browser receives metadata and lifecycle controls, not stored credential plaintext. See Bring your own provider credentials for the security boundary.

In strict BYOK mode, a request without a resolvable workspace provider credential fails with the typed no-provider-connected error. The broker does not fabricate a model response or silently use a shared ambient credential on that path.

What's next

• Making requests — verified chat, streaming, embeddings, and models contracts
• Authentication — API keys, workspace boundaries, and provider credentials
• Errors — layer-specific envelopes and recovery decisions