Most AI coding agents assume your code lives on GitHub. A lot of it doesn't. Here's what it takes to run gated AI changes on the Git host you already use.
If you read the docs of most AI coding agents closely, one line keeps showing up: works with repositories hosted on GitHub. That's a reasonable place to start. GitHub is huge.
But plenty of teams aren't there. Enterprises run self-managed GitLab. Agencies live in Bitbucket. Privacy-minded teams and a growing slice of open source run Gitea or Forgejo on their own servers. For them, "just use the agent" means moving their code first, which usually means not using the agent.
We built AI CodeOps to work where the code already is.
One workflow, five Git hosts
AI CodeOps connects to GitHub, GitLab, Bitbucket, Gitea and Forgejo. The flow is the same on each:
- Connect once. Use OAuth for the hosted services, or a personal access token for your own server. For a self-hosted GitLab, Gitea or Forgejo, you enter the server's public https address and paste a token
- Describe a change in the browser. Agents plan it, write it, review it and run the project's own tests in a sandbox
-
Get a pull request on your host. When the change passes the release gate, AI CodeOps pushes a branch named
codeops/run-<id>and opens a pull request (a merge request on GitLab) into the default branch
A team that uses more than one host can run all of them from one workspace.
What it does, and what it doesn't
For teams that own their tooling, the second list matters as much as the first. On Gitea and Forgejo, for example, AI CodeOps:
- Does list the repositories, branches, files and pull requests the connected account can see
- Does push a branch and open a pull request, with commits authored as "AI CodeOps"
- Doesn't need admin access. An account that can read the repository, push a branch and open pull requests is enough
- Doesn't install webhooks, post commit statuses or comment on your pull requests
- Never pushes to your default branch directly
Tokens are encrypted at rest and never written into a repository's Git configuration. To remove access, disconnect in AI CodeOps or revoke the token on your server.
The gate goes with you
The point isn't only that the agent can reach your host. It's that the same rules apply there. A change opens a pull request only when your tests passed on that exact version, the AI review of the real diff has no open high or critical findings, and a person approved it. If a check errors, the gate fails closed.
A pull request opened by AI CodeOps can be merged from the console only after the release gate passes again. You can also merge it on your host as usual, and if you want those merges gated too, your CI can ask the gate for a verdict through an API and fail the job on a "no".
A note for Codeberg users
Codeberg runs Forgejo, so the connection works with https://codeberg.org and a personal access token. But read Codeberg's Terms of Use first: they don't allow projects that mostly consist of code written by generative AI tools. That's their call, and worth respecting.
Try it on your host
AI CodeOps runs in the browser, with nothing to install. It's free during the open beta, and free forever for non-commercial open-source projects.
- Integrations: aicodeops.dev/integrations
- Open source program: aicodeops.dev/pricing/open-source
Drafted with AI assistance and edited by the AI CodeOps team.
Top comments (0)