Galpi
git push from a chat
When a repository's credentials can write, the chat's agent pushes through Galpi. The credentials stay on the server.
The agent makes commits inside the VM and pushes through Galpi. The VM pushes to Galpi with its chat's token; Galpi checks the rules (branch, person, author) and then pushes to the real repository. The VM never holds write credentials.
Turning it on
- In Project settings → Repository, open the repository and press Check write access under git push. When it passes you can turn on Allow pushes from chats. Each repository is separate.
- Set the branch pattern people may push to (default
galpi/{person}/*) and whether force pushes are allowed, per repository. - On Members, turn pushing on for the people who should. The default is read-only, and external users can't be turned on. If "members may push" was on when the project was created, invited members can push too.
Built-in repositories (started empty) can be pushed to from the start: any branch including the default one, never by force.
In a chat
Tell the agent something like "commit this and push it to galpi/galpi-push skill), checks which branches it may push to, and pushes.
- Commits are made with the name and email of the person chatting. Galpi refuses a push that includes commits authored by someone else.
- Refused pushes are listed in the chat's push refusals with the reason (branch rule, permission, author).
- Update never drops commits made in the workspace; they are kept on
galpi-saved/….
Galpi stops at the push. Open the pull or merge request in GitHub, GitLab or Forgejo, or add a skill or MCP connection that lets the agent do it.
commit · push · git · upload code · save to the repository