> For the complete documentation index, see [llms.txt](https://docs.klai.studio/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.klai.studio/reference/apis-and-services/streaming-proxy/bf-streaming-api-chat.md).

# Klai Streaming API (Chat)

{% hint style="info" %}
The BetterForms chat streaming endpoint is a lightweight streaming API for simple OpenAI and Gemini text generation.
{% endhint %}

## Overview

This endpoint still exists in the current BetterForms server and is registered at:

* `POST /stream/create`

It is a simpler streaming endpoint than `/llm/query`:

* supports `openai` and `gemini`
* streams basic `delta` and `end` events over BetterForms messaging
* does not expose the richer tool orchestration and cancellation contract documented for `/llm/query`

## Request Shape

| Key                        | Type      | Description                                                                                                                                        |
| -------------------------- | --------- | -------------------------------------------------------------------------------------------------------------------------------------------------- |
| `apiKey`                   | `string`  | BetterForms API key                                                                                                                                |
| `channels`                 | `array`   | BetterForms messaging channels used for streamed events                                                                                            |
| `actionName`               | `string`  | Named action that handles the streamed events                                                                                                      |
| `payload`                  | `object`  | Provider request payload                                                                                                                           |
| `payload.provider`         | `string`  | Provider name. Current implementation supports `openai` and `gemini`. Defaults to `openai` if omitted.                                             |
| `payload.apiKey`           | `string`  | API key for the selected provider                                                                                                                  |
| `payload.stream`           | `boolean` | If `true`, returns a streaming setup acknowledgment and emits events over messaging. If `false`, returns the result directly in the HTTP response. |
| `payload.messages`         | `array`   | Conversation messages                                                                                                                              |
| `payload.model`            | `string`  | Model name for the selected provider                                                                                                               |
| `payload.generationConfig` | `object`  | Optional Gemini-specific generation settings                                                                                                       |

## Example

```json
{
  "apiKey": "BFAPI_xxxxxxxx-xxxxxx-xxxxxx",
  "channels": [
    "anonymous"
  ],
  "actionName": "assistantReceiveResultsStream",
  "payload": {
    "provider": "openai", // or "gemini"
    "apiKey": "sk-xxxxxxx-xxxxxxxx-xxxxxxx", // Your OpenAI or Gemini API Key
    "stream": true,
    "messages": [
      {
        "content": "You are a helpful assistant.",
        "role": "system"
      },
      {
        "content": "Tell me a fun fact about space.",
        "role": "user"
      }
    ],
    "model": "gpt-4-turbo-preview", // or a Gemini model like "gemini-pro"
    // "generationConfig": { // Example for Gemini
    //   "temperature": 0.7,
    //   "maxOutputTokens": 2048
    // }
  }
}
```

## Streaming Behavior

When `payload.stream` is `true`, the endpoint returns a setup acknowledgment such as:

```json
{
  "success": true,
  "message": "Streaming setup complete"
}
```

The generated content is then delivered over BetterForms messaging.

Current event types for this endpoint are intentionally simple:

* `delta`
* `end`

Example streamed event:

```json
{
  "type": "delta",
  "content": "A streamed chunk of text"
}
```

```json
{
  "type": "end"
}
```

## Non-Streaming Behavior

When `payload.stream` is `false`, BetterForms returns the result directly in the HTTP response.

* OpenAI returns the provider response object
* Gemini returns a normalized response with `choices[0].message.content`

## Provider Notes

### OpenAI

* Uses the older chat-completions streaming flow in the current implementation
* Sends streamed text chunks as `delta` events

### Gemini

* Uses Gemini streaming directly
* Expects the last message to be a `user` message
* Supports `payload.generationConfig`

## Related Pages

* [Klai Streaming API (LLM Query)](/reference/apis-and-services/streaming-proxy/bf-streaming-api-llm-query.md)
* [Klai Streaming API (Assistants)](/reference/apis-and-services/streaming-proxy/bf-streaming-assistants-api.md)