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

# AI Evaluate (Stream)

> Stream an AI evaluation against the current page as Server-Sent Events in an OpenAI-compatible format.

## Overview

Same input as `POST /browser/sessions/{id}/ai/evaluate`, but the response is a long-lived Server-Sent Events stream. Response `Content-Type` is `text/event-stream`.

Each chunk follows the OpenAI Chat Completions streaming shape — a `data:` line carrying a JSON object, separated by a blank line. The stream terminates with `data: [DONE]`.

## Path Parameters

<ParamField path="id" type="string" required={true}>
  Browser session ID (UUID).
</ParamField>

## Body

Same shape as the non-streaming endpoint.

<ParamField body="prompt" type="string" required={true}>
  Natural-language instruction for the model.
</ParamField>

<ParamField body="mode" type="string" default="simple">
  One of `simple`, `agent`.
</ParamField>

<ParamField body="maxSteps" type="integer" default="20">
  Agent-mode step cap.
</ParamField>

<ParamField body="model" type="string">
  Explicit LLM model ID.
</ParamField>

<ParamField body="schema" type="object">
  JSON schema for structured output.
</ParamField>

## Example Request

<CodeGroup>
  ```bash cURL theme={null}
  curl -N -X POST "https://api.scrapengine.io/api/v1/browser/sessions/baa3f390-fa6e-4a24-b84a-a575a5f3a9c7/ai/evaluate/stream" \
    -H "Authorization: Bearer $SCRAPENGINE_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{
      "prompt": "Summarize this page in 3 bullets.",
      "mode": "simple"
    }'
  ```
</CodeGroup>

## Response

### Success Response (200)

* `Content-Type: text/event-stream`
* `Cache-Control: no-cache`
* `Connection: keep-alive`

Each chunk is a JSON payload matching OpenAI's streaming format. Example stream:

```
data: {"id":"agent-baa3f390","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"role":"assistant"}}]}

data: {"id":"agent-baa3f390","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":"- Example"}}]}

data: {"id":"agent-baa3f390","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":" Domain is..."}}]}

data: {"id":"agent-baa3f390","object":"chat.completion.chunk","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

data: [DONE]
```

On error, a final chunk is emitted with `delta.content: "Error: ..."` and `finish_reason: "stop"`, followed by `data: [DONE]`.

### Error Responses

| Status | Description                                                    |
| ------ | -------------------------------------------------------------- |
| `400`  | Invalid body — same validations as the non-streaming endpoint. |
| `401`  | Unauthorized — invalid or missing API key.                     |
| `404`  | Session not found or not owned by the caller.                  |

## Notes

* Use `-N` (no-buffer) with cURL to see chunks as they arrive.
* Any OpenAI-compatible SSE parser can consume this endpoint.
* For one-shot JSON output without streaming, use [`POST /browser/sessions/{id}/ai/evaluate`](/api-reference/endpoint/ai-evaluate).