# LLM Gateway (/llm-gateway)



## Why use it [#why-use-it]

Model providers use different credentials, model names, and request behavior. Without a gateway, those differences leak into product code and every provider change becomes integration work.

LLM Gateway gives each project one managed key and an OpenAI-compatible interface across OpenAI, Anthropic, and Gemini. Keep the client stable, then choose a provider model or a gateway alias with its own fallback chain in the `model` field.

```mermaid
flowchart LR
  C[Project code] -->|one key| G[LLM Gateway]
  G --> O[OpenAI]
  G --> A[Anthropic]
  G --> M[Gemini]
```

<Accordions>
  <Accordion title="Under the hood: project integration">
    The Agents template's project defaults declare the `llm_gateway` standard integration, so project creation adds it. Provider credentials live only in the gateway, never in the project. The gateway tags every call with the project and key that made it, then records token usage and cost per project.
  </Accordion>
</Accordions>

<Cards>
  <Card title="How to use it" href="/llm-gateway/how-to-use-it/">
    Call the gateway with an OpenAI-compatible client and pick a provider model or gateway alias.
  </Card>

  <Card title="API keys" href="/llm-gateway/api-keys/">
    See how deployments, workspaces, and local development get gateway credentials.
  </Card>
</Cards>