quote should be nullable in MessageContentTextAnnotationsFileCitationObject

[davidmigloz]

The quote property is marked as required, but the server doesn't include them in all the cases.

openai-openapi/openapi.yaml

Lines 11362 to 11397 in 893ba52

	MessageContentTextAnnotationsFileCitationObject:
	title: File citation
	type: object
	description: A citation within the message that points to a specific quote from a specific File associated with the assistant or the message. Generated when the assistant uses the "file_search" tool to search files.
	properties:
	type:
	description: Always `file_citation`.
	type: string
	enum: ["file_citation"]
	text:
	description: The text in the message content that needs to be replaced.
	type: string
	file_citation:
	type: object
	properties:
	file_id:
	description: The ID of the specific File the citation is from.
	type: string
	quote:
	description: The specific quote in the file.
	type: string
	required:
	- file_id
	- quote
	start_index:
	type: integer
	minimum: 0
	end_index:
	type: integer
	minimum: 0
	required:
	- type
	- text
	- file_citation
	- start_index
	- end_index

Example of response that doesn't include it:

{
  "id": "msg_Fn6zTxxx",
  "object": "thread.message",
  "created_at": 1715820384,
  "assistant_id": "asst_K4iArkxxx",
  "thread_id": "thread_3N31Erp0eCVhAhKh5BdMk1aQ",
  "run_id": "run_icH57L9xxx",
  "status": "completed",
  "incomplete_details": null,
  "incomplete_at": null,
  "completed_at": 1715820391,
  "role": "assistant",
  "content": [
    {
      "type": "text",
      "text": {
        "value": "...",
        "annotations": [
          {
            "type": "file_citation",
            "text": "【6:0†source】",
            "start_index": 1109,
            "end_index": 1121,
            "file_citation": {
              "file_id": "file-TrXXpxxx"
            }
          }
        ]
      }
    }
  ],
  "attachments": [],
  "metadata": {}
}

[dahifi]

Jumping in here. We did a fair amount of work in v1 to get this rendering well, and it seems like the quote is no longer returned for any of the v2 endpoints. Some of the older models might still do it, but not the new ones.

There are a couple threads on the community server, including this one with response from OAI staff member that it's not supported but "we'll work on" it.

Not sure where else to track it, but this OpenAPI repo seems more appropriate than the python API one.

But yes, I can also imagine that this causes _strict_response_validation to fail on every call.

[nknj]

Apologies for the oversight here -- we just rolled out a fix to remove quote from the openapi spec and will roll out updated sdks shortly.

[dahifi]

Apologies for the oversight here -- we just rolled out a fix to remove quote from the openapi spec and will roll out updated sdks shortly.

Not sure if this is the place to discuss this further, but having the quote available at the client level is going to be crucial for RAG validation. Removing it is a bad move.

[Adham-Mamedov]

@nknj Is team going to add quotes back?
It was really helpful in v1. I am working with legal documents, so having direct quotes would decrease the chance of assistant misinterpretation and errors.

[DennisPeeters]

Indeed, same issue

[sheldonrampton]

The 'quote' attribute is referenced in the OpenAI documentation:

https://platform.openai.com/docs/assistants/deep-dive/managing-threads-and-messages

If it has been removed, the documentation needs to be updated.

[eirikora]

In current version, I seem to be experiencing the same issue. See:
openai/openai-python#1498 (comment)