Vailabel Studio

Local-first, AI-assisted data labeling — a cross-platform desktop studio for images, video, text & audio.

Download · Features · Quick start · Architecture · Contributing

Overview

Vailabel Studio is an open-source desktop annotation tool in the spirit of LabelMe and Label Studio — but it runs entirely on your machine. Your data lives in a local SQLite database, images are referenced in place (no base64 copies), and everything works offline. On top of that familiar labeling workflow it adds an AI layer: one-click auto-labeling, click-to-segment, and an on-device assistant.

It's built as a native desktop app with Tauri (Rust) + React, so you get a small, fast binary with real filesystem access instead of a browser tab.

✨ Features

Label anything

Images — bounding boxes, polygons, keypoints, lines, polylines, circles, and free-draw.
Video — frame extraction, scene detection, and object labeling across time (FFmpeg pipeline).
Text — named-entity spans, relations, classification, Q&A, and translation.
Audio — waveform region labeling, diarization, and transcription.
Custom — define your own projects with Label-Studio-style JSON configs; a

config-driven editor renders the right tools automatically.

AI on top

Auto-label — run a YOLO detection model on the current image and review the suggestions.
Smart Segment — click an object to segment it with SAM (Segment Anything).
Copilot (experimental) — an offline, on-device vision assistant for label suggestions.

Built for real datasets

Local-first — a single SQLite database; fully offline, your data never leaves your machine.
Reference in place — point at a folder of images; nothing is copied or duplicated.
Import / export — LabelMe sidecars, JSONL, and dataset exports for model training.
Cloud storage (optional) — connect S3, Azure Blob, or Google Cloud Storage; credentials stay in your OS keychain.
Label-Studio-style workspace — a file queue, center canvas with a tool rail, and a docked

Classes / Regions / AI / Copilot panel.

Project status: Vailabel Studio is under active development. Core image/text/audio/video

labeling and YOLO/SAM assistance are usable today; the on-device Copilot and embedded training

runtime are evolving — expect rough edges and breaking changes.

⬇️ Download

Grab the latest installer for your platform from the Releases page:

| Platform | Installer | | --- | --- | | Windows | .exe (setup) or .msi | | macOS | .dmg — Apple Silicon (aarch64) and Intel (x64) | | Linux | .AppImage, .deb, or .rpm |

macOS builds are unsigned for now — if Gatekeeper blocks the app, right-click → Open the

first time (or xattr -dr com.apple.quarantine /Applications/Vailabel\ Studio.app).

🛠️ Development

Prerequisites

Node.js 20+ and Yarn (Classic / v1)
Run nvm use from the repository root to match the project's Node.js version
Rust (stable toolchain) — install via rustup
The platform build dependencies for Tauri v2 — see the

Tauri prerequisites guide

Run it

# install workspace dependencies
yarn install
run the full desktop app (Tauri + React)
yarn dev          # alias: yarn desktop
run just the React UI in the browser (no Rust backend)
yarn studio
tests, lint, typecheck
yarn test
yarn lint
yarn typecheck
build production bundles
yarn build        # web + studio frontend
yarn release      # native desktop installers (tauri build)

🧱 Project structure

This is a Yarn-workspaces monorepo:

apps/
├── studio/                # the Tauri desktop app
│   ├── src/               # React frontend — feature-sliced
│   │   ├── app/           #   router (lazy routes), providers, layout shell
│   │   ├── features/      #   studio · projects · ai-models · labels · …
│   │   └── shared/        #   ui (shadcn-style) · lib · services · ipc · types
│   └── src-tauri/         # Rust backend — Tauri commands + DDD modular crates
│       ├── crates/        #   per-domain crates (pure domain + Diesel/SQLite)
│       └── migrations/    #   the canonical SQLite schema
└── web/                   # marketing / documentation site

🏗️ Architecture

Frontend — React 19 + Vite + TypeScript + Tailwind v4, organized feature-first

(app/ + features/ + shared/); routes are code-split with React.lazy.

Desktop shell — Tauri v2; the React UI talks to Rust over a typed IPC bridge.
Backend — Rust as a DDD modular monolith: each domain is its own crate with a pure

core and a typed Diesel repository over a shared SQLite connection. Embedded migrations are the single source of truth for the schema.

🤝 Contributing

Contributions are welcome! Please read CONTRIBUTE.md to get set up, and our Code of Conduct before opening a pull request. Found a security issue? See SECURITY.md — please don't file it as a public issue.