DocFinder

Local-first semantic search for your documents.
Supports PDF, Word, PowerPoint, OpenDocument, HTML, EPUB, Markdown, plain text, and more.
Everything runs on your machine — no cloud, no accounts, complete privacy.

Features

• Semantic search — find documents by meaning, not just keywords (PDF, DOCX, PPTX, ODT, HTML, EPUB + others)
• AI chat — ask questions about any document and get precise answers, powered by local Qwen3.5 models (automatically selects the best model for your hardware)
• Bring your own models (Ollama) — connect to a local or remote Ollama server and pick any embedding or chat model it serves
• 100% privacy mode — per-indexing-run checkbox: files marked private never leave your machine (local embedder required, remote LLMs blocked for chat)
• 100% local — your files never leave your machine
• GPU accelerated — auto-detects Apple Silicon (Metal), NVIDIA (CUDA), AMD (ROCm)
• Cross-platform — native apps for macOS, Windows, and Linux
• Global shortcut (Experimental) — bring DocFinder to front from anywhere with a configurable hotkey

Download

Platform	Installer
macOS	DocFinder-macOS.dmg
Windows	DocFinder-Windows-Setup.exe
Linux	DocFinder-Linux-x86_64.AppImage

macOS — open the DMG, drag DocFinder to Applications, then right-click → Open on first launch (Gatekeeper warning — normal for unsigned open-source apps).

Windows — run the installer; if SmartScreen appears choose More info → Run anyway.

Linux

chmod +x DocFinder-Linux-x86_64.AppImage && ./DocFinder-Linux-x86_64.AppImage

Run from Source

Requires Python 3.10+ and make.

git clone https://github.com/filippostanghellini/DocFinder.git
cd DocFinder
make setup   # create .venv and install all dependencies
make run     # desktop GUI
make run-web # web interface at http://127.0.0.1:8000

Runtime acceleration (auto)

DocFinder automatically selects the best available runtime on your machine:

• NVIDIA: ONNX CUDA provider when available, otherwise PyTorch CUDA
• AMD: ONNX ROCm provider when available, otherwise PyTorch fallback
• Apple Silicon: PyTorch MPS when available (fastest), ONNX ARM64 fallback otherwise
• Intel Mac / CPU-only: ONNX or PyTorch CPU fallback

Indexing uses an adaptive parallel parser strategy by default, selected automatically based on your machine resources.

Custom models via Ollama

DocFinder is not locked to its built-in models. In Settings → Models & Ollama you can connect to any Ollama server and choose which models to use:

• Server URL (e.g. http://127.0.0.1:11434) plus an optional API key for remote Ollama providers, with a one-click Test connection that lists the installed models
• Embedding model — pick any embedding model served by Ollama; note that changing the embedding model invalidates the existing index, so use the Re-index now button afterwards
• Chat model — pick any chat model served by Ollama for the AI Chat; applies immediately, no reindex needed

Both choices are independent, so you can run a small embedding model and a larger chat LLM side by side on the same server. The server can be local or a VPS/remote Ollama provider — embeddings and chat requests go over HTTP.

100% privacy mode

The 100% privacy checkbox in the Index tab marks an indexing run as strictly local:

• Indexing requires a local embedding model — the checkbox is disabled while a remote Ollama server is configured
• AI chat on privacy-marked documents refuses remote LLMs (Ollama on localhost is fine — data never leaves your machine); the built-in GGUF models are always allowed
• Privacy-marked documents show a shield badge in the Documents tab, and re-indexing preserves the flag

Documents indexed without the flag keep working with any model, including remote Ollama servers.

Contributing

Contributions are welcome, feel free to open an issue or submit a pull request.

License

Licensed under the GNU Affero General Public License v3.0 (AGPL-3.0).

DocFinder was originally released under the MIT License. Starting from version 1.1.1 the license was changed to AGPL-3.0 to comply with the PyMuPDF licensing requirements, as PyMuPDF itself is AGPL-3.0 licensed.