Agent Queue
A background service that runs AI coding-agent CLIs against a developer's own git repositories, giving each task an isolated worktree, a terminal session, a recorded outcome, and a policy-driven path toward the default branch.
A user describes work as tasks; Agent Queue decides which task is ready, gives it an isolated git worktree, starts a coding-agent CLI inside a terminal session, records what came back, and carries the finished branch toward the default branch under a chosen policy. Tasks, dependencies, gates, sessions and outcomes are stored in PostgreSQL rather than in a single model's context, so a crashed or exhausted agent produces a retry instead of ending the job.
It is aimed at people who already run coding-agent CLIs by hand and want a queue, isolation and an audit trail around them, rather than a hosted product; the underlying idea is a local software factory that expresses work, assigns it to a fleet of agents, and keeps it moving.
Installation is a single script for macOS or Windows via WSL2 that sets up a checkout and a command-line tool, and hands remaining setup to a resumable onboarding wizard covering prerequisites, PostgreSQL, agent CLIs, configuration, the daemon and a local dashboard. It is written in Python and MIT licensed.