DocuBrowse: A Local, Offline AI Search Engine for Your Document Pile

DocuBrowse is an open-source desktop tool that indexes a scattered collection of files — PDFs, Word docs, spreadsheets, presentations, ebooks, HTML, and plain text — into a knowledge base you can query by meaning rather than exact keywords. It combines SQLite FTS5 full-text search with semantic similarity from local embeddings, defaulting to a hybrid mode that weights results 70% semantic and 30% keyword. Clicking a result generates a book-jacket-style synopsis on demand, cached after first use, so you can gauge a document before opening it.

The defining feature is that everything runs on-device. All AI work is handled locally through Ollama, using nomic-embed-text for embeddings and dolphin3 for summaries, which means no internet connection, no accounts, no API keys, and no per-query costs. It also ships a post-ingest PII scanner that detects patterns like Social Security numbers, credit cards, bank account details, and passport numbers, then removes and permanently blacklists any matching files — a privacy posture aimed at keeping sensitive documents out of the index entirely.

Engineering details point at a tool built for real-world messiness: pdfplumber with a pypdf fallback for malformed PDFs, OCR routing for scanned image-only files, Calibre-backed handling of MOBI and AZW formats, parallel core-aware extraction, and memory-safe worker limits. Typical search latency is under 150ms. It’s packaged for Linux (RPM, DEB, tarball), Windows, and macOS at version 0.9.0, positioning it as a private, self-hosted alternative to cloud document search and RAG services.