chrisryugj/Docufinder

Anything – 100 % local document search engine

What it is – A desktop application (Windows & macOS Apple Silicon) that indexes the files on your PC and lets you find them by their contents instead of just their filenames. It works completely offline for the core search functions and optionally plugs into LLM services (Gemini, OpenAI‑compatible servers such as Ollama, vLLM, LiteLLM) for natural‑language Q&A and document summarisation.

Key capabilities

Feature	How it works
Full‑text search	Registers folders, builds a SQLite FTS5 index and a vector index (usearch HNSW). Queries return matches in milliseconds, even for thousands of files.
Filename search	In‑memory cache gives “Everything‑style” instant lookup of file names, with sortable column view.
AI Q&A (optional)	Sends the user’s natural‑language question plus the most relevant text chunks to a Gemini or OpenAI‑compatible endpoint. The answer is shown with source document and page number, and can be saved as Markdown.
AI summarisation (optional)	Right‑click a file → Summarise; the selected LLM produces a concise summary appropriate to the document type.
Real‑time sync	File system watcher updates the index automatically when files are added, edited or removed.
Preview & layout view	Click a result to open a preview. Supports true layout rendering for HWP/HWPX (via rhwp), PDF (page images) and DOCX/PPTX/XLSX. Password‑protected HWP/HWPX can be opened by entering the password in the preview.
Export	Save search results or whole documents as Markdown; drag‑and‑drop files to other apps; compare two versions of the same document side‑by‑side.
OCR & formula detection	Optional OCR (PaddleOCR) extracts text from scanned PDFs and images; LaTeX formulas can be recognised when the OCR option is enabled.
Lite build	A “network‑code‑free” Windows‑only binary for air‑gapped environments (no AI, OCR, or auto‑update).

Supported formats – HWP/HWPX (including old HWP3), DOCX, PPTX, XLSX/XLS, PDF, common image types, TXT/MD, and EML mail files. Korean language processing uses Lindera morphological analysis; embeddings are generated with a KoSimCSE‑roberta ONNX model.

Installation

• Windows – Download the Anything_<ver>_x64-setup.exe (≈380 MB). If the machine lacks Microsoft Edge WebView2, also download the MicrosoftEdgeWebView2RuntimeInstallerX64.exe and install it first. A “Lite” installer (Anything.Lite_…_x64-setup.exe) is provided for highly restricted corporate PCs.
• macOS (Apple Silicon) – Download the .dmg, drag to Applications, then open via right‑click → Open (or clear the quarantine attribute with xattr). Only Apple‑Silicon is supported; Intel Macs are not.
• Minimum hardware: 8 GB RAM (16 GB recommended), ≥1 GB free disk space. All core parsers, OCR models and the search engine run locally; no background download is required after the first run (except optional AI/semantic models, which are ~106 MB each).

AI integration

• Enable in Settings → AI and supply an API key for either Google Gemini or any OpenAI‑compatible endpoint (including self‑hosted LLMs via Ollama, vLLM, LiteLLM, etc.).
• When disabled, the app behaves as a pure local search tool – no network traffic leaves the machine.
• The only outbound traffic (when AI is on) is the user’s question and the selected text chunks; responses are returned from the configured LLM.

Architecture snapshot

React 19 + TypeScript + Tailwind   ← UI
Tauri 2 IPC                         ← bridge
Rust backend (clean architecture)   ← parsers, indexer, search engine
  ├─ Parsers: hwpx (kordoc), docx, pdf/ocr, txt, eml …
  ├─ Indexer: SQLite FTS5 (keyword), usearch HNSW (semantic)
  └─ Search: keyword + hybrid RRF merging
SQLite + ONNX runtime (embeddings, OCR) stored locally

License – Business Source License 1.1 (free for non‑production use; commercial deployment requires a paid license). After 15 April 2030 the code will automatically re‑license to Apache 2.0. Third‑party components (PDFium, ONNX Runtime, PaddleOCR, Node.js, kordoc, etc.) are listed in THIRD_PARTY_NOTICES.md.

Who might benefit?

• Professionals who need to locate information inside large collections of Korean documents (e.g., contracts, reports, meeting minutes) without uploading anything to the cloud.
• Companies with strict data‑security policies that still want optional AI‑assisted Q&A on internal documents.
• Power users who appreciate a Windows‑native, Electron‑alternative desktop UI with rich preview, export and diff features.

---

All details above are taken directly from the repository’s README; no additional features have been inferred.