productcode & review
git push reasonos feat/retry-budget
Git hosting that survives agents. Ten thousand pushes a day.
Repositories live on sharded data planes with backups, LFS and branch rules. Push over HTTPS or SSH, browse, blame and search in the hub, and every push is durable before it is acknowledged.
how you use it
Hosted git, in order.
01
Move in with one push
Add the remote and push. Your workspace address is the remote; your SSH key is the credential. A mirror keeps browsing, blame and search fast in the hub even when no branch server is running.
02
Every push is durable at the ack
Each receive gets a ticket. The pre-receive hook records the ref with it before git is allowed to move, so a crash between the two cannot lose a branch. LFS uploads are written through to backup before they are acknowledged.
03
The agent’s loop: pushes at volume
Agents push far more often than people. Every push gets the same durable ack and the same CI and QA runs, a janitor sweeps the lock files git leaves behind at that volume, and geometric repacks and commit graphs keep the repository fast under it. Branch rules do not care who pushed.
04
Protect main with rules, not conventions
Required approvals, required checks, dismiss stale approvals, require the last push approved, thread resolution, linear history, signed commits. A merge happens server-side after a clean merge-tree, under a compare-and-swap on the base, so a racing push aborts instead of being overwritten. Repos are assigned a shard when created and never rehashed, so adding capacity never moves anyone.
not built yet: connecting a GitHub or GitLab repository directly; today import is a plain git remote · renaming or archiving a project from settings
The full reference lives in the docs.
works with
Open a branch and try it.
Boots in seconds. Your team joins with a link.