Connect clients
Complete a first request before configuring other clients. Use a user API key, separate from provider credentials and Admin integration keys.
Choose the client protocol
Section titled “Choose the client protocol”| Client | Address or interface | Requirements |
|---|---|---|
| OpenAI-compatible SDK, IDE or CLI | Base URL: http://localhost:8787/v1 | User key, gateway model ID and an active Chat or Responses surface |
| Anthropic-compatible client | Request URL: http://localhost:8787/v1/messages | A Messages surface; configure the SDK base according to its URL joining rules |
| Native Gemini client | http://localhost:8787/v1beta/models/model-id:generateContent | A Gemini surface and native request body |
| MCP client | http://localhost:8787/mcp or /mcp/server-slug | An enabled MCP server and a user key or OAuth |
Replace local addresses with your deployed Proxy URL. Some clients append /v1 themselves; inspect the final path to avoid repeating it.
Models and route groups
Section titled “Models and route groups”Clients use the gateway’s model ID. The route maps it to the upstream model name. Specify a group as model-id:route-group, for example your-model:free. Available protocols depend on the group’s active surfaces.
Authenticated GET /v1/models returns LLMs by default. Use kind=image, kind=asr, kind=tts or kind=embedding for other types; kind=all includes all types. The compatible kind=audio filter includes ASR and TTS.
Test Responses
Section titled “Test Responses”curl -sS http://localhost:8787/v1/responses \ -H 'Authorization: Bearer sk-your-api-key' \ -H 'Content-Type: application/json' \ -d '{"model":"your-route-model","input":"Hello"}'A working Chat route does not automatically provide Responses, Messages or Gemini. Check the request surfaces, then inspect request logs.
See client examples for additional requests and fields.