AI PDF Chatbot GitHub 2026: 7 Open Source Projects to Self-Host
Want to chat with PDFs without sending documents to a cloud? These seven GitHub projects let you self-host the whole stack.
💡 What You Will Learn
Want to chat with PDFs without sending documents to a cloud? These seven GitHub projects let you self-host the whole stack.
📜 Table of Contents
Why Self-Host a PDF Chatbot
Three reasons: privacy (contracts, medical records, unreleased docs), cost (free at scale beyond your hardware), and control (no vendor changing features under you). The projects below are the ones with real traction in 2026 - stars fetched 2026-08-14.
The Projects
AnythingLLM (64,687 stars) - the easiest entry. One Docker command, connect Ollama or any API, drag PDFs in, chat. Desktop apps for Windows/Mac/Linux too.
RAGFlow (87,912 stars) - the parsing heavyweight. Its document understanding (deep doc parsing, tables, OCR, layouts) is the best of the bunch for messy PDFs.
private-gpt (57,431 stars) - privacy-first, everything local, no telemetry. Fewer features, strongest guarantee.
Onyx (Danswer) (31,582 stars) - knowledge-base oriented: connect PDFs plus Slack, Notion, Google Drive, wikis. Enterprise connectors and permissions make it the team pick.
Khoj (36,480 stars) - personal AI for all your files (PDF, markdown, org). Chat, semantic search, automated research.
txtai (13,000 stars) - the embeddable library: not an app but a Python/Rust toolkit to build a PDF chatbot into your own product.
Open WebUI (148,695 stars) - the general local ChatGPT UI with built-in document RAG. If you already use Ollama, this adds chat-with-PDFs in minutes.
The Comparison Matrix
| Project | Stars | Best for | Setup |
|---|---|---|---|
| AnythingLLM | 64,687 | Fastest start | 1 command |
| RAGFlow | 87,912 | Messy/scanned PDFs | Docker compose |
| private-gpt | 57,431 | Strict privacy | pip install |
| Onyx | 31,582 | Team knowledge base | Docker compose |
| Khoj | 36,480 | Personal files | Docker/app |
| txtai | 13,000 | Building your own | pip install |
| Open WebUI | 148,695 | Ollama users | Docker |
The Recommendation
Start with AnythingLLM. If your PDFs are scans or table-heavy, move to RAGFlow. If the whole team needs access with permissions, Onyx. All of them run with a local model (Ollama) for total privacy.
FAQ
Which project needs the least technical skill? AnythingLLM - desktop app, no Docker required.
Can these run on a laptop? Yes - with a small model (7-8B quantized) and CPU, answers are slow but work; 16GB+ RAM recommended.
Do they support scanned PDFs? RAGFlow leads on OCR; others have basic OCR at best.
Can I connect a cloud API instead of local models? Yes - all of them support OpenAI-compatible APIs, so you can mix local and cloud.
❓ FAQ
Which project needs the least technical skill?
AnythingLLM - desktop app, no Docker required.
Can these run on a laptop?
Yes - with a small model (7-8B quantized) and CPU, answers are slow but work; 16GB+ RAM recommended.
Do they support scanned PDFs?
RAGFlow leads on OCR; others have basic OCR at best.
Can I connect a cloud API instead of local models?
Yes - all of them support OpenAI-compatible APIs, so you can mix local and cloud.
Written by our editorial team; tools listed here are tested or verified against public sources. Links point to official sites or GitHub repos for reference only โ no paid placements.
