An AI-assisted dataset annotation platform that runs entirely within your tailnet — private, fast, and built for teams.
SPUR Founder Track winner at Hack Canada 2026! Check out the Devpost here :-) https://devpost.com/software/datascale
One of our team members was once responsible for annotating sensitive medical imaging research data. This is a product of a realization that there is too much friction between maintaining dataset privacy and team annotator collaboration.
- AI-Assisted Annotation — Leverage an integrated AI service to auto-suggest labels, dramatically speeding up the annotation workflow
- Private by Design — Deployed entirely on your Tailnet; no data leaves your private network
- Role-Based Access — Tailscale ACL policies gate access per user, keeping annotators, reviewers, and admins in their lanes
- Real-Time Collaboration — Multiple annotators can work simultaneously across the same dataset
- Export Ready — Annotated datasets can be exported in standard formats for immediate use in model training pipelines
- Node.js v18+
- Python 3.10+
- Tailscale installed and authenticated on all machines
- A Tailscale auth key (for service-to-service communication on the tailnet)
git clone https://github.com/t9nzin/datascale.git
cd datascaleApply the example ACL policy to your Tailnet via the Tailscale admin console:
# Reference the example policy included in the repo
cat acl_policy_example.jsonEnsure all services (client, server, ai-service) are joined to the same tailnet before proceeding.
cd ai-service
pip install -r requirements.txt
python main.pycd server
npm install
npm run devcd client
npm install
npm run devThe app will be accessible at the client's Tailscale IP address. Only devices on your tailnet can reach it.
| Layer | Technology |
|---|---|
| Frontend | React, Vite, Zustand |
| Backend | Node.js, Express, SQLite |
| AI Service | FastAPI, MobileSAM, SAM2, Ollama, YOLO-World |
| Infrastrucure | Tailscale Serve, Tailscale ACLs |