ChatGPTNextWeb
[Feature Request] 能否加入通过插件分析PDF的功能
🥰 Feature Description
加入通过插件分析PDF的功能
🧐 Proposed Solution
用过插件形式
📝 Additional Information
No response
Bot detected the issue body's language is not English, translate it automatically.
Title: [Feature Request] Is it possible to add the function of analyzing PDF through plug-in?
这个应该也可以通过包装fastgpt的api实现（这里是接口文档地址）： https://doc.tryfastgpt.ai/docs/development/openapi/dataset/#%e6%95%b0%e6%8d%ae
实现逻辑
- 更新https://github.com/ChatGPTNextWeb/NextChat-Awesome-Plugins/tree/main/plugins/fastgpt 这里的schema，增加创建知识库，以及
创建外部文件库集合的api - 编辑mask，增加context内容：告诉LLM，当收到一个（或者多个）PDF的链接的时候，先创建知识库，并将PDF作为集合导入到知识库，并记住知识库的datasetId
- 用户提问的时候，尝试调用fastgpt的搜索测试接口查询相关内容。再由大模型总结回答。
Bot detected the issue body's language is not English, translate it automatically.
This should also be achieved by wrapping the fastgpt API (here is the interface document address): https://doc.tryfastgpt.ai/docs/development/openapi/dataset/#%e6%95%b0%e6%8d%ae
Implement logic
- Update the schema here at https://github.com/ChatGPTNextWeb/NextChat-Awesome-Plugins/tree/main/plugins/fastgpt, and add the API to create a knowledge base and create an external file library collection.
- Edit the mask and add context content: Tell LLM that when receiving a link to one (or multiple) PDFs, first create a knowledge base, import the PDF as a collection into the knowledge base, and remember the datasetId of the knowledge base.
- When the user asks a question, try to call fastgpt’s search test interface to query related content. Then the large model summarizes the answer.
Bot detected the issue body's language is not English, translate it automatically.
#5384 If the plugin runtime is later implemented to support uploading files, the previously mentioned step of creating a collection can be used Create a file collection This API
这个应该也可以通过包装fastgpt的api实现（这里是接口文档地址）： https://doc.tryfastgpt.ai/docs/development/openapi/dataset/#%e6%95%b0%e6%8d%ae
实现逻辑
- 更新https://github.com/ChatGPTNextWeb/NextChat-Awesome-Plugins/tree/main/plugins/fastgpt 这里的schema，增加创建知识库，以及
创建外部文件库集合的api- 编辑mask，增加context内容：告诉LLM，当收到一个（或者多个）PDF的链接的时候，先创建知识库，并将PDF作为集合导入到知识库，并记住知识库的datasetId
- 用户提问的时候，尝试调用fastgpt的搜索测试接口查询相关内容。再由大模型总结回答。
好的，感谢，我研究一下
Bot detected the issue body's language is not English, translate it automatically.
This should also be achieved by wrapping the fastgpt API (here is the interface document address): https://doc.tryfastgpt.ai/docs/development/openapi/dataset/#%e6%95%b0%e6%8d%ae
Implement logic
- Update the schema here at https://github.com/ChatGPTNextWeb/NextChat-Awesome-Plugins/tree/main/plugins/fastgpt, and add the API to create a knowledge base and create an external file library collection
- Edit the mask and add context content: tell LLM that when receiving a link to one (or more) PDFs, first create a knowledge base, import the PDFs as a collection into the knowledge base, and remember the datasetId of the knowledge base.
- When the user asks a question, try to call fastgpt's search test interface to query related content. Then the large model summarizes the answer.
Okay, thanks, I'll look into it
这个应该也可以通过包装fastgpt的api实现（这里是接口文档地址）： https://doc.tryfastgpt.ai/docs/development/openapi/dataset/#%e6%95%b0%e6%8d%ae
实现逻辑
- 更新https://github.com/ChatGPTNextWeb/NextChat-Awesome-Plugins/tree/main/plugins/fastgpt 这里的schema，增加创建知识库，以及
创建外部文件库集合的api- 编辑mask，增加context内容：告诉LLM，当收到一个（或者多个）PDF的链接的时候，先创建知识库，并将PDF作为集合导入到知识库，并记住知识库的datasetId
- 用户提问的时候，尝试调用fastgpt的搜索测试接口查询相关内容。再由大模型总结回答。
Bot detected the issue body's language is not English, translate it automatically.
This should also be achieved by wrapping the fastgpt API (here is the interface document address): https://doc.tryfastgpt.ai/docs/development/openapi/dataset/#%e6%95%b0%e6%8d%ae
Implement logic
- Update the schema here at https://github.com/ChatGPTNextWeb/NextChat-Awesome-Plugins/tree/main/plugins/fastgpt, and add the API to create a knowledge base and create an external file library collection
- Edit the mask and add context content: tell LLM that when receiving a link to one (or more) PDFs, first create a knowledge base, import the PDFs as a collection into the knowledge base, and remember the datasetId of the knowledge base.
- When the user asks a question, try to call fastgpt's search test interface to query related content. Then the large model summarizes the answer.
https://chatgpt.com/g/g-V2KIUZSj0-pdf-ai-pdf
这里有一个由myaidrive.com提供的使用人数比较多的GPTs，名字叫做PDF AI PDF，这一个GPTs使用的Action是这个schema，只不过这个action只支持oauth登录，如果以某个方式拿到oauth之后的user_access_token然后，就可以将这个当作插件放到NextChat内部使用。
但是，我只有myaidrive的free帐号，不知道怎么拿到可以调用api的apikey
{
"openapi": "3.1.0",
"info": {
"title": "AI Drive",
"description": "Super-fast, interactive chats with PDFs of any size, complete with page references for fact checking.",
"version": "v0.0.1"
},
"servers": [
{
"url": "https://aipdf.myaidrive.com",
"description": "Production server"
}
],
"paths": {
"/summarize_pdf": {
"post": {
"description": "Provide the text content of the PDF linked for summarization. Users can specify page range as an option.ALWAYS PROVIDE QUOTES AND PAGE CITATIONS.",
"operationId": "summarize",
"parameters": [],
"requestBody": {
"content": {
"application/json": {
"schema": {
"$ref": "#/components/schemas/summarizeRequestSchema"
}
}
},
"required": true
},
"responses": {
"200": {
"description": "Successful response"
}
},
"deprecated": false,
"security": [],
"x-openai-isConsequential": false
}
},
"/upload_and_search_pdf": {
"post": {
"description": "Semantic query into a URL link to a document. THINK STEP BY STEP. ALWAYS PROVIDE QUOTES AND PAGE CITIATIONS. BREAK COMPLEX QUESTIONS INTO SEVERAL QUERIES.",
"operationId": "upload_and_search_pdf",
"parameters": [],
"requestBody": {
"content": {
"application/json": {
"schema": {
"$ref": "#/components/schemas/upload_and_search_pdfRequestSchema"
}
}
},
"required": true
},
"responses": {
"200": {
"description": "Successful response"
}
},
"deprecated": false,
"security": [],
"x-openai-isConsequential": false
}
},
"/create_pdf_map": {
"post": {
"description": "Create a PDF map for the file the user uploaded to the GPT",
"operationId": "createPDFMap",
"parameters": [],
"requestBody": {
"content": {
"application/json": {
"schema": {
"$ref": "#/components/schemas/create_pdf_mapRequestSchema"
}
}
},
"required": true
},
"responses": {
"200": {
"description": "Successful response"
},
"500": {
"description": "Unable send file(s)."
}
},
"deprecated": false,
"security": [],
"x-openai-isConsequential": false
}
}
},
"components": {
"schemas": {
"upload_and_search_pdfRequestSchema": {
"properties": {
"pdf_url": {
"type": "string",
"title": "pdf_url",
"description": "url where the PDF file is stored. e.g. https://myaidrive.com/eHrCoFs6i23ygrxL/Democracy_in.pdf"
},
"query": {
"type": "string",
"title": "query",
"description": "Query to perform semantic search."
}
},
"type": "object",
"required": [
"pdf_url",
"query"
],
"title": "upload_and_search_pdfRequestSchema"
},
"summarizeRequestSchema": {
"properties": {
"pdf_url": {
"type": "string",
"title": "pdf_url",
"description": "url where the PDF file is stored. e.g. https://myaidrive.com/eHrCoFs6i23ygrxL/Democracy_in.pdf"
},
"start_page": {
"type": "integer",
"title": "start_page",
"description": "The starting page number for the summary. Optional.",
"minimum": 1,
"example": 1
},
"end_page": {
"type": "integer",
"title": "end_page",
"description": "The ending page number for the summary. Optional.",
"minimum": 1,
"example": 20
}
},
"type": "object",
"required": [
"pdf_url"
],
"title": "summarizeRequestSchema"
},
"create_pdf_mapRequestSchema": {
"properties": {
"openaiFileIdRefs": {
"type": "array",
"items": {
"type": "string",
"description": "List of files to send to BE."
}
}
},
"type": "object",
"required": [
"openaiFileIdRefs"
]
}
},
"securitySchemes": {
"apiKey": {
"type": "apiKey"
}
}
}
}
@Why-SVIP
I find one GPTs, named PDF AI Reader Chat (4.1 ★), and get action import to NextChat.
then can talk with PDF url:
https://github.com/ChatGPTNextWeb/NextChat-Awesome-Plugins/tree/main/plugins/chatpdf