PDF-GPT
Ranked in AI PDF Assistants ·No free plan on record
About
PDF-GPT is an open-source application for asking questions about PDF content using GPT functionality. It accepts an uploaded document or a PDF URL, divides the document into smaller chunks and creates embeddings with a Deep Averaging Network Encoder. Semantic search selects relevant content, with a documented retrieval flow that uses KNN to fetch the top five chunks before sending results to OpenAI for an answer. Answers can include the PDF page number in square brackets. The project describes a local Gradio playground available in a browser at localhost:7860, an API exposed with langchain-serve, Docker Compose use and deployment to Jina Cloud. Its API documentation includes ask_url and ask_file endpoints, while sample requests pass an OPENAI_API_KEY through request environment variables. The README says the Universal Sentence Encoder download is 915 MB if fetched at runtime. The project is licensed under the MIT License, and its maintainer is seeking voluntary contributors to help maintain the application and work through backlog items.
Who it is for
PDF-GPT may suit developers who want a self-hosted PDF question-answering app with a local browser playground or API endpoints. It also fits teams willing to work with OpenAI credentials and an open-source project seeking volunteer maintenance help.
What is good
- Accepts uploaded PDFs or PDF URLs.
- Answers can include PDF page citations.
- Includes ask_url and ask_file API endpoints.
- Documents local, Docker Compose and Jina Cloud use.
- Licensed under the MIT License.
What to know first
- Runtime Universal Sentence Encoder download is 915 MB.
- Sample API requests require an OPENAI_API_KEY.
- Maintainer seeks voluntary contributors for maintenance.
Verdict
PDF-GPT offers PDF ingestion, semantic retrieval and page-number citations, with documented local and deployment options. Account for the model download and OpenAI key requirements when planning an implementation.
Compared on AI PDF assistants
- Source citations
- Yesgithub.com
- Access platforms
- Web, API, self_hostedgithub.com
Facts
- PDF chat
- PDF-GPT lets users chat with an uploaded PDF using GPT functionality.github.com · 4 Oct 2026
- Document processing
- The app splits documents into smaller chunks and generates embeddings with a Deep Averaging Network Encoder.github.com · 4 Oct 2026
- Semantic search
- It searches PDF content semantically and sends the most relevant embeddings to OpenAI.github.com · 4 Oct 2026
- Page citations
- Answers can include the PDF page number in square brackets.github.com · 4 Oct 2026
- Input options
- The documented workflow accepts an uploaded PDF or a PDF URL.github.com · 4 Oct 2026
- API
- The README describes exposing the app as an API with langchain-serve, including ask_url and ask_file endpoints.github.com · 4 Oct 2026
- Local use
- The README documents a local Gradio playground available in a browser at localhost:7860.github.com · 4 Oct 2026
- Deployment
- The project documents Docker Compose use and deployment to Jina Cloud.github.com · 4 Oct 2026
- API key
- The sample API requests pass an OPENAI_API_KEY in the request environment variables.github.com · 4 Oct 2026
- Retrieval limit
- The documented flow retrieves the top five chunks with KNN before generating an answer.github.com · 4 Oct 2026
- Encoder download
- The README says the Universal Sentence Encoder download is 915 MB if fetched at runtime.github.com · 4 Oct 2026
- License
- The project states that it is licensed under the MIT License.github.com · 4 Oct 2026
- Maintainer appeal
- The README says the maintainer is seeking open-source contributors to help maintain the application and take backlog items voluntarily.github.com · 4 Oct 2026
Best PDF-GPT alternatives
See all 20
Where it ranks on Inferse
- Best AI PDF Assistants in 2026#9 of 27
Sources
- github.com/jasontobeagile/PDF-GPT· checked 4 Oct 2026