03 — Product · AI model & dataset hub
Gitsi
Code, weights and data in one git repo, served to the Hugging Face tools you already use. The home for AI models, datasets and code — built in India.

$ curl -fsSL https://www.gitsi.com/cli.sh | sh$ git checkout -b fine-tune-hindi$ git push origin fine-tune-hindiremote: Create a new pull request for 'fine-tune-hindi'
The problem
- GitHub holds the code.
- Hugging Face holds the weights.
Our approach
- Gitsi keeps both in one repository.
- A team can review, reproduce and ship from one place.
- Served to the Hugging Face tools you already use.
Features
What Gitsi does.
01
Branch a model like you branch code
Training scripts, configs, checkpoints and the model card live in one repository. Large files go through Git LFS, so a clone stays fast and every file has a history.
02
Pull requests for models and data
Push a branch and open a pull request — review models the way you review code.
03
Hugging Face-compatible API
Set HF_ENDPOINT and keep your tools: huggingface_hub, transformers and datasets download, upload and stream as they do today. Plain git works too.
04
Upload big weights from a browser tab
Chunked, resumable, up to 5 GB a file, straight into Git LFS. A dropped connection resumes where it stopped.
05
Scanned on upload, quarantined if bad
Uploads are scanned, and anything bad is quarantined before others can pull it.

01 — Branch a model like you branch code
Specifications
Built to spec.
API
Hugging Face-compatible — set HF_ENDPOINT
Storage
Git + LFS · up to 5 GB a file from the browser
Tested with
huggingface_hub 0.28–2.0 · transformers 5.17 · datasets 5.0 · git 2.43 · git-lfs 3.4
Deployment
Hosted, or self-hosted on-premises / air-gapped
Pricing
Public work free · Pro USD 1/month
Status
Free during the private alpha
How it works
One system, clearly layered.
Who it's for
Built for teams that depend on it.
Students and researchers
₹0 to start, forever for public work. Support in IST working hours.
Teams
One organization for your code, checkpoints and datasets, with roles that say who can read, write and merge.
Regulated labs
The same Gitsi, on your servers, on-premises or air-gapped, for organizations whose models and data must stay put.

