OfflineFind · 拾文

Offline file & full-text search for PDF, Markdown and DOCX.

Remember the words. Find the file.

简体中文 · Releases · User guide · Development

OfflineFind automatically discovers local files for offline desktop search. Open it and type: no folder selection is required. Find images, archives, code and folders by name; search PDF, Markdown and DOCX contents with highlighted excerpts. The interface retains its frosted-glass appearance.

Stable release: v0.1.1. Supports automatic local discovery and offline search within the documented limits. See release notes and limitations.

What works

• Automatically discover local disks, index file/folder names first, and extract document contents independently.
• Choose Names + contents, File names only, or Document contents only.
• Literal phrase queries, space-separated AND conditions, type/folder filters and bookmarks.
• Real excerpts with highlights; PDF page, Markdown line and DOCX paragraph references.
• Local SQLite FTS5 bigram index, followed by exact text verification to reject false positives.
• Read-only isolated parsers, periodic disk reconciliation, pause/resume, editable subfolder exclusions and clear-data controls.
• Paginated problem-file lists and individual retries that stay pending across pause/restart.
• Frameless window with integrated minimize/maximize/close, drag regions and edge resizing.
• Light/dark/system appearance, optional frosted glass, Chinese/English interface and keyboard navigation.
• Manual update checks, platform-matched downloads, progress/cancellation and SHA-256 verification in Settings.
• One running instance per index; reopening activates its window. Linux portable bundles include a user installer.
• No account, telemetry, external fonts, remote document rendering or automatic updates.

Download

Get the package for your platform from GitHub Releases: Windows x64 ZIP, macOS Apple Silicon ZIP, or Linux x64 DEB / tar.gz. Extract the entire archive before launching OfflineFind. Windows/macOS packages include Python; Linux reuses the system runtime.

• Windows 11 x64: uses Microsoft Edge WebView2. The runtime must already be installed; OfflineFind does not silently download it. Most Windows 11 installations include it. Air-gapped machines need the runtime provisioned separately.
• macOS 14+ Apple Silicon: uses the system WebKit. The application is not notarized and has no developer signing certificate. macOS may require approval through System Settings → Privacy & Security after the first launch attempt.
• These binaries are unsigned (macOS packaging may apply an ad-hoc signature). Verify the published SHA-256 checksum. Do not disable system-wide security settings.
• Ubuntu 22.04 / 24.04 x64: install the .deb through your software installer or sudo apt install ./OfflineFind-*-linux-amd64.deb. Search OfflineFind in the application menu. The portable tar.gz runs with ./OfflineFind/offlinefind; see the Linux guide. Both use system Python 3.10+, GTK 3 and WebKit; Python modules are included, so no pip setup is needed.

First search

1. Open OfflineFind; local file discovery starts automatically.
2. Type immediately. Names become searchable first, followed by document contents.
3. Search for 部署, budget approval, or "offline deployment".
4. Select a result to read its extracted text. Use Open original for the original layout.

Ctrl+K / Cmd+K focuses search inside the app. Names support single-character terms; content terms need at least two characters. Space-separated terms must all appear in the same document. Search is literal, not AI-powered or semantic.

Privacy and limits

Files are never moved, renamed or modified. The local index contains extracted document text and is not separately encrypted. Protect it as you would the original documents, using OS account permissions and disk encryption. Use Index & exclusions to exclude folders, or clear local data and pause indexing.

Automatic indexing skips virtual filesystems, network mounts, cache/dependency directories, symlinks and its own index. Inaccessible directories are skipped without elevation. Windows discovers drive-letter fixed disks; Linux/macOS traverse local directory trees. Changes are reconciled at approximately 30-second intervals plus scan time; NTFS MFT / USN acceleration is not implemented.

Scanned pages need OCR, which is not included. Password-protected files, unsupported encodings, parser errors and size limits are shown explicitly. File names may remain searchable when their contents cannot be parsed. Default limits: 100 MiB per file, 1,000 PDF pages and 10 MiB of extracted text. DOCX previews contain ordinary paragraphs and table text, not original Word pagination.

Full desktop memory and some searches exceed the original PRD budgets; see the measurements below. See the roadmap.

Run from source

Requires Python 3.10+ (Windows/macOS bundles use Python 3.12; Linux uses system Python).

git clone https://github.com/asoming/offlinefind.git
cd offlinefind
python -m venv .venv
# Windows: .venv\Scripts\activate
source .venv/bin/activate
python -m pip install -e '.[dev]'
offlinefind

Development-only browser mode (bound to loopback, authenticated per process):

offlinefind --serve --port 8765 --folder /absolute/path/to/documents

--folder limits development tests to an explicit scope; normal launching needs no such argument. File opening uses the desktop app. Browser mode is a development convenience, not a hosted service. Indexing and searching work offline. Manual update checks/downloads connect to GitHub without sending local documents, paths or indexes; installing source dependencies requires a prepared package cache or internet access.

Contributing and license

Read the development guide and contributing notes. Please use synthetic or explicitly shareable examples in bug reports; do not upload private documents or indexes.

MIT © 2026 asoming. See LICENSE and third-party notices.

Performance measurements

The acceptance benchmark covers 10,000 mixed documents, 100 queries and one million synthetic names. Full local desktop idle RSS was 574 MiB; some document queries took 0.4–0.8 seconds, and a broad million-name query took 2.7 seconds. Native desktop checks run on Windows, macOS and Linux. Read the measurements, raw reports and remaining limits; cold-cache and original low-memory targets are not claimed.

Renamed from Shiwen

OfflineFind is the English product name; 拾文 remains the Chinese name. The repository is now asoming/offlinefind. Upgrade normally: existing indexes, bookmarks and exclusions are preserved. The shiwen Python module, Debian package ID, application identity and existing data directories remain for compatibility; offlinefind is the primary command. Shiwen-* release files are byte-identical compatibility aliases for older updaters. New downloads use OfflineFind-*.