Client Setup
Applications reach the LLM Gateway with the official SDK they already use. Set the SDK's base URL to the gateway, and keep the rest of the code, including the API key, as it is.
| SDK | Base URL | API key header | Guide |
|---|---|---|---|
| OpenAI | http://<gateway-host>:<port>/v1 | Authorization: Bearer <key> | OpenAI SDK |
| Anthropic | http://<gateway-host>:<port> | x-api-key: <key> | Anthropic SDK |
The port is the one the tcp · server in front of the matching llm · server listens on. Which provider answers depends on the model in the request and the routes on llm · proxy, not on the SDK.
Which API Key to Send
With credential pass-through, Zilla forwards the key the SDK sends to the provider that serves the request. Send the key for that provider: an OpenAI key for a gpt-* model, an Anthropic key for a claude-* model, whichever SDK you use.
Errors
Zilla returns errors in the error format of the SDK's API, whether they come from the provider or from Zilla, such as an LLM guardrail rejection or a failed authorization.