vector-watcher

Vector Watcher is a cross-platform desktop application for inspecting and managing LanceDB vector databases through a graphical interface. It targets users who work with local or remote vector embeddings and want to browse, query and organise them without writing scripts. ## What it does - **Connection manager** — configure, persist and switch between multiple local or remote LanceDB instances. - **Schema and table inspector** — browse vector datasets, structural schemas, metadata and row counts. - **Vector similarity search** — run distance queries using L2, Cosine or Dot Product metrics directly from the UI. - **Zero-setup sidecar** — the local Python backend is auto-managed and runs on an isolated loopback socket at `127.0.0.1:8765`. ## Architecture The project pairs a Tauri native host (Rust) with a React/TypeScript webview front end. The Tauri core manages a bundled Python/FastAPI engine launched as a sidecar process via Uvicorn, which in turn communicates with LanceDB Core (stated version 0.30.2) through native bindings. Communication between the webview and the native host uses local webview IPC; the backend is bound locally rather than exposed on the network. ## Installation Pre-compiled binaries are published on the project's Releases page. Documented targets include Linux (Ubuntu 24.04+, x86_64/arm64, `.AppImage` and `.deb`), macOS (Apple Silicon and Intel, `.dmg`) and Windows 11 (x64, `.msi` and setup `.exe`). ## Development The repository is a monorepo split into a `frontend` and a `backend` directory. Prerequisites listed are Node.js v20+ LTS, a recent stable Rust toolchain for Tauri, and Python 3.11+ with Poetry. Setup steps cover `npm install` in the frontend, `poetry install` plus a start script for the Python sidecar, and running the Tauri dev command. Standalone installers are produced by running the backend build script and then a platform-specific `tauri:build` script. ## Documentation and community Additional material — user guide and operations, architecture and FastAPI protocol notes, and packaging/binary sidecar workflow — is kept in the project's GitHub Wiki. The project is maintained by Keshav Mohta and Dhwani Khandelwal, welcomes issues and pull requests, and is distributed under the MIT License. ## Notes This overview reflects only what the README documents: a desktop GUI client for LanceDB. Performance characteristics, dataset size limits and the full feature set beyond the listed capabilities are not described in the source material.