Protocols and compatibility
Each API configuration can expose one or more Lumoswitch Runtime unified client formats. Requests go directly to the Lumoswitch inference service and authenticate with the same Lumoswitch access key.
| Format | Method | Routes |
|---|---|---|
| OpenAI | GET | /v1/models |
| OpenAI | POST | /v1/chat/completions, /v1/responses, /v1/embeddings |
| Anthropic | POST | /v1/messages, /v1/messages/count_tokens |
| Gemini | POST | /v1beta/models/{model}:generateContent, :streamGenerateContent, :countTokens |
| Gemini | POST | /models/{model}:generateContent, :streamGenerateContent, :countTokens |
Use https://api.lumoswitch.com/v1 for OpenAI clients and https://api.lumoswitch.com for Anthropic or Gemini clients. Every failover candidate must support the capabilities used by the client.
Use the Lumoswitch access key in the client's native authentication header: Authorization: Bearer for OpenAI, x-api-key for Anthropic, or x-goog-api-key for Gemini. Never put an upstream provider key in a downstream client.
These are Lumoswitch Runtime's unified translation routes: an Anthropic- or Gemini-shaped client can target any compatible configured upstream. Native provider pass-through paths such as /anthropic/*, /gemini/*, and /cohere/* remain private because they do not use the configuration's model and credential boundary.