A developer built a $0 merge queue to keep parallel AI coding agents from crashing each other's builds

When four or five AI coding agents share one 8GB MacBook Air, the first casualty isn't the model; it's the build. Pushing, compiling, and testing in parallel turns the machine into one shared race track, and the most common finish is a force quit.

A [Show HN post from developer funador](https://news.ycombinator.com/item?id=49104747) addresses that coordination failure with a local FIFO merge queue for Claude Code, Anthropic's CLI for AI-assisted coding. The tool gives each agent its own numbered worktree lane, then funnels every push through a machine-wide lock so two agents never build, test, or land at the same time. A land command rebases the lane onto the integration branch and pushes through the FIFO queue.

The author reports running 4–5 parallel agents pushing about 90 commits a day on the same laptop, and wanted to avoid paying GitHub Actions minutes for roughly 90 daily pushes. The project's README frames it as a $0, no-PR-required alternative to GitHub's hosted Merge Queue, which is gated to Enterprise Cloud for private repos and bills per build minute. Same idea: serialize landings, test before merge, keep history clean, but run locally instead of in someone else's cloud.

The trade-off is explicit: the default mode is direct rebase and push with no pull-request review, and checksRequired: false is a supported, named state. For a solo dev running many agents on one laptop, that cost-control posture is the point. For a team, the missing review lane is the obvious next question.