Skip to content
SPECIFICTION
Get listed

PDF-ChatGPT

github.com

WebWindowsmacOSLinuxAPISelf-hosted
Visit github.com
Spec score 7.4 #22 of 27 in AI PDF Assistants

What it is

PDF-ChatGPT is a free, open-source, self-hosted application for asking natural-language questions about PDF files and receiving answers with document and page citations. It can index multiple PDFs and show document status, page counts, and chunk counts. Semantic search uses a persistent FAISS index, with a relevance threshold and optional per-document scoping. Answers come from OpenAI’s Responses API and can stream token by token; SQLite keeps conversation history for follow-up questions. The browser interface renders Markdown, offers dark and light themes, and adapts to mobile and tablet layouts. Uploaded PDFs, the index, and conversation history are stored locally, but text is sent to OpenAI for embeddings and chat generation. An OpenAI API key with billing enabled is required, and token use is metered. Uploads must be PDFs under 25 MB; scanned PDFs without extractable text need OCR, which is not included. The default setup is single-tenant and has no login, so the maker advises against public exposure without authentication or network restrictions. It runs locally or on documented hosting platforms, with persistent storage needed across redeploys.

Spec sheet

Purpose
PDF-ChatGPT lets users ask natural-language questions about PDFs and receive answers grounded in document content with page citations.Read 8 Oct 2026 · github.com
Deployment
The README describes local operation and deployment with platforms such as Render, Railway, Fly.io, or Azure App Service, with persistent storage needed to retain data across redeploys.Read 8 Oct 2026 · github.com
Integrations
Out of the box, the project uses OpenAI embeddings and the OpenAI Responses API; the maker says adapting it to another LLM provider would require code changes.Read 7 Oct 2026 · github.com
License
The repository states that the project is released under the MIT License.Read 8 Oct 2026 · github.com
Access control
The default setup is single-tenant with no login system, and the maker advises against exposing it publicly without authentication or network restrictions.Read 7 Oct 2026 · github.com
Answer generation
The app uses OpenAI's Responses API and can stream answers token by token.Read 7 Oct 2026 · github.com
API and setup help
FastAPI provides autogenerated Swagger, ReDoc, and OpenAPI documentation, and the repository includes a beginner's walkthrough.Read 8 Oct 2026 · github.com
Chat features
Conversation history is stored in SQLite, prior turns are used for follow-up questions, and answers can stream token by token over Server-Sent Events.Read 8 Oct 2026 · github.com
Citations
Answers include citations to the source document and page.Read 7 Oct 2026 · github.com
Conversation features
Conversation history is stored in SQLite and prior turns are used for follow-up questions.Read 7 Oct 2026 · github.com
Data handling
The maker says PDFs, the FAISS index, and conversation history are stored locally, while text is sent to OpenAI for embeddings and chat generation.Read 7 Oct 2026 · github.com
Document library
It supports uploading and indexing multiple PDFs and shows document status, page counts, and chunk counts.Read 7 Oct 2026 · github.com
File limits
Uploads are restricted to PDF files under a configurable size limit, and scanned PDFs without extractable text are not supported without adding OCR.Read 7 Oct 2026 · github.com
Integration
The app requires an OpenAI API key and uses OpenAI embeddings; the README says API usage is metered per token and requires billing to be enabled.Read 8 Oct 2026 · github.com
Interface
The web interface includes Markdown rendering, code syntax highlighting, dark and light themes, and a responsive layout for mobile and tablet.Read 8 Oct 2026 · github.com
Notable limits
Scanned image PDFs without an extractable text layer are unsupported without adding OCR, and the README lists OCR as a future improvement.Read 8 Oct 2026 · github.com
PDF library
It supports uploading and indexing multiple PDFs and shows document status, page counts, and chunk counts in a sidebar library.Read 8 Oct 2026 · github.com
RAG stack
The application is built with FastAPI, LangChain, FAISS, and the OpenAI Responses API.Read 8 Oct 2026 · github.com
Search
Semantic search supports top-k retrieval, a relevance threshold, and optional per-document scoping.Read 8 Oct 2026 · github.com
Security setup
The README advises configuring the OpenAI key and other settings through environment variables or deployment secrets, and recommends TLS termination through Nginx or Caddy.Read 8 Oct 2026 · github.com
Storage
Uploaded PDFs and the FAISS index are stored locally, while conversation and message history is persisted in SQLite.Read 8 Oct 2026 · github.com
Usage costs
Users need an OpenAI API key with billing enabled, and embedding and chat usage is metered per token.Read 7 Oct 2026 · github.com

Strong points and limits

Strong points

  • Answers cite the source document and page.
  • Supports multiple PDFs per chat.
  • Search can be scoped to one document.
  • MIT-licensed and open source.

Limits

  • OpenAI token usage is metered.
  • Scanned PDFs need OCR added.
  • Default setup has no login.
  • Uploads must be under 25 MB.

Who it suits

It suits people who want a self-hosted PDF question-and-answer interface and can configure an OpenAI API key and deployment security. Scanned documents require a separate OCR addition.

Verdict

PDF-ChatGPT provides citations and follow-up conversations over indexed PDFs, while keeping files, index, and history local. OpenAI billing is still required, and the no-login default calls for care before exposing an instance publicly.

Where Specifiction places it

Alternatives to PDF-ChatGPTAll alternatives →