> ## Documentation Index
> Fetch the complete documentation index at: https://docs.neuraltrust.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Chat

> Use OpenAI, Anthropic, Cohere, or Gemini request formats while TrustGate selects and adapts the upstream provider.

TrustGate accepts several provider-shaped chat APIs on the same consumer. The endpoint you
call defines the request and response format your application uses; it does not force
TrustGate to route to that provider.

```text theme={null}
/{consumer_slug}/{fixed_route}
```

For example, an Anthropic Messages request can be routed to an OpenAI registry. TrustGate
adapts the request, response, and supported streaming events while preserving the client
format.

## Endpoints and client formats

| Method | Path after consumer slug                       | Client format                |
| ------ | ---------------------------------------------- | ---------------------------- |
| `POST` | `/v1/chat/completions`                         | OpenAI Chat Completions      |
| `POST` | `/v1/messages`                                 | Anthropic Messages           |
| `POST` | `/v1/responses`                                | OpenAI Responses             |
| `POST` | `/v2/chat`                                     | Cohere Chat v2               |
| `POST` | `/v1beta/models/{model}:generateContent`       | Gemini generateContent       |
| `POST` | `/v1beta/models/{model}:streamGenerateContent` | Gemini streamGenerateContent |

Only these fixed shapes are accepted. Similar but unlisted paths, such as
`/v2/chat/completions`, return **404**.

## Send a request

Use the **LLM URL** and credentials from the consumer **Connect** tab:

```bash theme={null}
curl -X POST "https://<llm-host>/<consumer-slug>/v1/chat/completions" \
  -H "X-AG-API-Key: <consumer-api-key>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "auto",
    "messages": [{"role": "user", "content": "Explain zero trust in one sentence."}]
  }'
```

TrustGate accepts `X-AG-API-Key`, `x-api-key`, or
`Authorization: Bearer ag_…`. OAuth2 and OIDC bearer tokens are also accepted when
configured. Private deployments may require `X-AG-Gateway-Slug`.

See [Auth](/trustgate/concepts/auth) and [Quickstart](/trustgate/getting-started/quickstart).

## Model selection and routing

The consumer controls which registries and models can serve the request:

* Use `"model": "auto"` for load balancing or Smart routing.
* Use an allowed model ID to request a specific model.
* Gemini puts the model in the URL path instead of the request body.
* Model allowlists and defaults are enforced before TrustGate calls the provider.

Routing references are resolved by TrustGate and are not forwarded upstream. Successful
responses include `X-Selected-Provider` and `X-Selected-Model`, which are useful when
debugging routing decisions.

See [Model resolution](/trustgate/routing/model-resolution),
[Load balancing](/trustgate/routing/load-balancing), and
[Smart routing](/trustgate/routing/smart-routing).

## Streaming

For OpenAI, Anthropic, Cohere, and Responses clients, set `"stream": true` in the request.
For Gemini, call `:streamGenerateContent` or add `?alt=sse`.

TrustGate adapts normal server-sent events to the format expected by the client. If an
upstream stream stops unexpectedly after the HTTP response has started, TrustGate emits
this fixed error event; the already-sent HTTP status remains `200`:

```text theme={null}
data: {"error":{"message":"upstream stream terminated unexpectedly","type":"upstream_error"}}
```

<Note>
  Provider-specific features can differ. TrustGate adapts supported request and response
  fields, but does not guarantee every provider extension is portable across every provider.
</Note>