siGit Code & Deploy

siGit is Git hosting for the AI era. Version your repositories, publish open-weights models with full model cards and LFS-backed files, and hand the keys to your AI agent over the Agent Client Protocol. Code and models live side by side, in the same storage and the same calm UI.

Why this exists

We kept running into the same split. Source code lived in one place, model weights lived somewhere else, and the two never really met. Pushing a model meant a different tool, a different account, and weights scattered across services.

So we built siGit. It is plain Git underneath, so nothing about your workflow has to change, but it treats a multi-gigabyte model the same way it treats a source file: something you can host, browse, and version without fuss.

What it does

• Host public and private repositories with fast diffs and deep file browsing
• Clone, fetch, and push over Git with token-authenticated HTTPS
• Publish open-weights models with model cards parsed straight from your README
• Store large weights with Git LFS, and see sizes, formats, and quants at a glance
• Connect AI coding agents over the Agent Client Protocol
• Use the JSON API and the siGit Code & Deploy desktop and mobile apps
• Deploy through smbCloud

Who makes it

siGit is built by Splitfire AB, a small software company in Sweden. Authentication, email, and deploys run on the smbCloud platform. We use siGit ourselves daily, which keeps the bug list honest.

Questions or feedback: splitfire@setoelkahfi.se.