Agent Query Stream

Stream a response to a query in a simple markdown format.

This endpoint provides a simplified streaming response:

First, it streams the ID mapping for source linking
Then it streams raw markdown content as it comes from the LLM

The client only needs to concatenate the markdown chunks to get the full document, and can use the ID mapping to resolve citations to their sources.

Request body:

query: The query to ask the agent (required)
tickers: Optional list of ticker symbols to filter documents by
sourceType: Optional source type(s) to include (available options: '10-K', '10-Q', '8-K', '20-F', '40-F', '6-K', 'transcript', 'DEF 14A', etc.). Can be a single string or a list of strings.
startDate: Optional start date for document filtering (format: YYYY-MM-DD)
endDate: Optional end date for document filtering (format: YYYY-MM-DD)
sourceLink: Optional source link(s) to search within. Can be a single string or a list of strings. No other documents will be queried by the agent but these. Use to restrict the search to user-selected documents.

Returns a streaming response with these event types:

'id_mapping': A mapping of IDs to source links for citation resolution

{
  "type": "id_mapping",
  "mapping": {
    "#uuid1234": {
      "sourceLink": "http://rest-api.captide.co/document?id=12345678",
      "sourceMetadata": {
        "sourceType": "10-K",
        "ticker": "AAPL",
        "date": "2024-01-01",
        "fiscalPeriod": "FY 2023",
        "documentId": "12345678"
      }
    }
  }
}

'markdown_chunk': A chunk of markdown text

{
  "type": "markdown_chunk",
  "content": "This is a part of the markdown response with a citation [#uuid1234]."
}

'warning': Notification when requested information is not found

{
  "type": "warning",
  "message": "Could not find information about X in the provided documents."
}

'done': Signals the end of the stream
```
{
  "type": "done"
}
```

'error': Error information if something goes wrong

{
  "type": "error",
  "detail": "An error occurred while processing the request."
}

Client implementation recommendations:

First process the 'id_mapping' event to obtain source information for citations
Then accumulate all 'markdown_chunk' events to build the complete response
Display warnings to the user when 'warning' events are received
When a 'done' event is received, the response is complete
Use the ID mapping to create clickable citations in the rendered markdown

Integration with document search: The 'sourceLink' parameter can be used in conjunction with the document search endpoint to:

Let users search and select specific documents they want to analyze
Then pass those document IDs to this endpoint to ensure the AI only references those selected documents
This creates a more user-controlled research experience where they can direct the AI to specific filings

Note: All JSON fields use camelCase naming convention.

Request Body — REQUIRED
`query` Query — REQUIRED The query to ask the agent
`tickers` Tickers List of ticker symbols to filter documents by (e.g., ['AAPL', 'MSFT']). Check the '/companies' endpoint for available tickers.
`sourceType` Sourcetype Source type(s) to include. Available options: 10-K, 10-Q, 8-K, 20-F, 40-F, 6-K, transcript, DEF 14A, DEFM14A, DEF 14C, DEFM14C. Can be a single string or a list of strings.
`start_date` Start Date Start date for document filtering (format: YYYY-MM-DD)
`end_date` End Date End date for document filtering (format: YYYY-MM-DD)
`sourceLink` Sourcelink Source link(s) to search within. No other documents will be queried by the agent but these. Can be a single string or a list of strings. Use to restrict the search to user-selected documents.

Responses

200

Successful Response

Schema — OPTIONAL
undefined

422

Validation Error

Schema — OPTIONAL

detail object[] — OPTIONAL

loc undefined[]

msg Message

type Error Type