Ask Question

Develop Env

https://dev.your-api-server.com

POST

/questions

Chat with the document, response in server-sent events mode by default, you can change it through stream parameter.

Consume one question quota whether it's upon single document or collection.

Response:

if document is pdf/doc/docx format, structure is as follows:

{
    "data": {
        "answer": "answer to the question",
        "id": question_id,
        "source_info": [
            {
                # key: page number
                # value: rects
                '0': [[38.1063, 557.8058, 553.9003, 584.0043]]
            },
            {'1': [[38.0, 152.3994, 523.6151, 178.6392]], 'upload_id': 'xxxx'},
            {'0': [[38.0, 758.0623, 537.0082, 784.0]], 'upload_id': 'xxxx'},
            ...
        ]
    }
}

if document is md/epub/txt/website format, structure is as follows:

{
    "data": {
        "answer": "answer to the question",
        "id": question_id,
        "source_info": [
            {
            # key: element data-index
            # value: element xpath
            198: [{xpath: "/html/body/div/div[199]"}]
            },
            {
            "material": "", # selected text with HTML tags
            "indexes": [
                3,
                4,
                5,
                6
            ],
            "focusNode": "div[1]/p[5]/text()[1]",
            "upload_id": "903d971d-8250-47cc-a649-4b6ca35032dc",
            "anchorNode": "div[1]/p[2]/text()[1]",
            "focusOffset": "225",
            "anchorOffset": "2"
        }
            ...
        ]
    }
}

answer: chunks of answer, may be Markdown format to support rich text, for example: Tables. For detailed_citation answer, the span tag chunk may be as follows:

autobiography captured the pre-Nazi Europe[<span data-index="0">1</span>]

The tag attr data-index is the index of the source in source_info array, which is used to highlight the source of the previous answer sentences in your PDF viewer. The highlighting method is same as source_info, and just use slice of source_info array as parameter.

id: id of the question, you can use it to GET /questions/{question_id} later.

Please note: you should store id in your database, because we don't have a GET /questions/list API to list all questions for now.

source_info: only responses in the last chunk of server-sent events mode, may be an empty list. So if the last chunk doesn't contain source_info attr, it means error occurred. Page number may not be ordered, you can use this information to highlight the source of specific upload_id document of the answer in your PDF viewer, by
calling drawSources method of our DOCViewerSDK, and converting source_info to Source parameter.

Request Example

Shell

JavaScript

Java

Swift

curl --location --request POST 'https://dev.your-api-server.com/questions' \
--header 'Content-Type: application/json' \
--data-raw '{
    "upload_id": "stringstringstringstringstringstring",
    "question": "string",
    "stream": true,
    "selected_meta": {
        "material": "string",
        "upload_id": "stringstringstringstringstringstring",
        "rects": [
            {
                "pageNumber": 1,
                "outline": [
                    1,
                    1,
                    1,
                    1
                ]
            }
        ]
    },
    "history": [],
    "search_entire_doc": true,
    "detailed_citation": false,
    "language": "string",
    "model_type": "gpt-3.5-turbo"
}'

Response Example

200 - Example 1

null

Request

Body Params application/json

upload_id

string

Upload Id

required

document id or collection id, created by POST /documents/upload or POST /collections APIs accordingly.

>= 36 characters<= 36 characters

question

string

Question

required

question to ask

>= 1 characters<= 3000 characters

stream

boolean

Stream

optional

If set, partial message deltas will be sent, like in ChatGPT of OpenAI. Tokens will be sent as data-only server-sent events as they become available.

Default:

true

selected_meta

Selected Meta

optional

When use Ask About Selected Text, this field is acquired from EVENT_TYPES.CHAT_ICON_CLICKED of our DOCViewerSDK, and then filled with the document upload_id, when ask upon a collection, it's the sub document id of collection.

Any of

SelectedMeta

HtmlSelectedMeta

material

string

Material

required

upload_id

string

Upload Id

required

>= 36 characters<= 36 characters

rects

array[object (PDFViewerRect) {2}]

Rects

required

>= 1 items

history

array[object (ChatMessage) {2}]

History

optional

chat history, used for Threaded Chat, this parameter simulates OpenAI' API. history is question and answer pairs, which is a list of ChatMessage objects. The max input tokens of history, selected_meta.material and question is 8000, which is a little less than OpenAI's limit of Embedding models. Sample parameter is as follows:

[
{"role": "user", "content": "Who..."},
{"role": "assistant", "content": "The..."},
{"role": "user", "content": "Where..."}
{"role": "assistant", "content": "The..."},
]

role

enum<string>

required

An enumeration.

Allowed values:

assistantuser

content

string

Content

required

search_entire_doc

boolean

Search Entire Doc

optional

if true: limit responses to current file's information, else: responses go freely with knowledge of our AI

Default:

true

detailed_citation

boolean

Detailed Citation

optional

Whether to show the source at the end of each sentence or not. Requires enabling the search_entire_doc setting.

Default:

false

language

string

Language

optional

preferred language of answer to the question, if not provided, we will try to detect the language automatically, which is flexible and also works well. You'd better use ISO language name, such as English, Chinese. But we can't promise, when AI can't answer your question well, it generally answers in English.

model_type

enum<string>

optional

Supported LLM enumeration.

Allowed values:

gpt-3.5-turbogpt-4gpt-4o

Default:

gpt-3.5-turbo

Examples

Responses

🟢200Successful Response

application/json

Body

No schema defined

🟠422Validation Error

Modified at 2024-07-26 06:57:23

Get Suggested Questions

Get Question