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

# Stream Chat

> Stream a chat response grounded in retrieved content.

The generated server embeds the latest user message, retrieves relevant chunks, and supplies them to the configured chat model. The response uses a server event stream.

### Body

<ParamField body="messages" type="array" required>
  Conversation messages with a `role` of `user`, `assistant`, or `system` and a string `content`.
</ParamField>

<ParamField body="topK" type="number">
  Maximum number of context chunks to retrieve. Defaults to the server's `TOP_K` value.
</ParamField>

### Events

Each SSE frame contains a JSON object in its `data` field.

| Type         | Fields  | Description                                              |
| ------------ | ------- | -------------------------------------------------------- |
| `text-delta` | `text`  | The next generated text fragment.                        |
| `done`       | `hits`  | Generation completed, with the retrieved context chunks. |
| `error`      | `error` | The stream failed after it started.                      |

```bash theme={null}
curl --no-buffer http://localhost:8080/chat \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $LARKUP_API_KEY" \
  -d '{
    "messages": [
      { "role": "user", "content": "How do I configure authentication?" }
    ],
    "topK": 4
  }'
```

```text theme={null}
event: message
data: {"type":"text-delta","text":"Configure "}

event: message
data: {"type":"text-delta","text":"SERVER_API_KEY."}

event: done
data: {"type":"done","hits":[...]}
```

Set `CHAT_API_KEY` and, optionally, `CHAT_MODEL` in the generated server environment. Custom models that use the OpenAI interface also use `CHAT_BASE_URL`.