CloudBlue Digital

AI coding assistants are transformative in an editor — but a backlog is not an editor. Hive is our orchestrator for running AI coding sessions against GitLab issues at the group level: every repo, one dashboard, sessions in parallel.

Labels as the control surface

Hive does nothing to an issue unless you label it. One label surfaces it on the dashboard for you to launch manually; another opts it into automation, where Hive's poller picks it up and starts a session on its own. It is a simple contract that keeps humans in charge of what gets automated, issue by issue.

Watch the work happen

Each session runs in an isolated working directory and streams its output live, so you can watch the model read code, run tests, and iterate — or ignore it entirely and check back when it is done. Sessions run in parallel across different issues and repositories.

Everything ends in review

The unit of output is a draft merge request. Nothing merges itself; your existing review culture is the quality gate. In practice that changes the shape of an engineering day: instead of starting work, you start reviews, and the backlog moves while you sleep.