A software factory for your AI coding agents.
Describe the work, let your agents build it in isolated sandboxes, and merge only the changes that pass your tests and an independent review. One pipeline — from a rough idea to a pull request you can trust.
$ brew install partyline-sh/tap/partylineAgents run on your hardware, in your git, with your keys — the server never holds a model key.
One agent in one terminal is a demo, not a team's workflow. The moment you point several agents at a real backlog, the questions are the same ones a factory answers: who decides what to build, where does the work happen so it doesn't collide, and what stops a plausible-but-wrong change from shipping? partyline is that factory — a pipeline that turns an idea into scoped tasks, builds each in isolation, and gates every change on evidence.
How it works
Describe
Tell the Requirements Agent what you want. It interviews you — product, technical, and UX decisions — then writes a plan of bounded, independently buildable tasks.
Plan
Review the breakdown as a team, on the board or in the CLI. Edit it, re-order it, and send only the tasks you approve to the backlog.
Build
Agents pick up tasks and build each one in its own isolated git worktree — in parallel, on your hardware, always off a fresh main so nothing collides.
Verify & ship
Every change runs your tests and an independent reviewer. What passes becomes a PR you merge; what fails is quarantined for a human — never shipped on trust.
Where it shines
Ship a feature as a team
Describe it together, watch the agents build it in parallel, and review the PRs that come back. Non-technical teammates work from the web; engineers stay in the terminal.
Turn a bug report into a merged fix
Paste the report, let an agent reproduce and fix it in a sandbox, and merge only after the tests it wrote — and an independent reviewer — both pass.
Put a fleet on the backlog
Send a stack of approved tasks to your agents at once. Each builds in its own worktree; you get a queue of PRs to review instead of one session to babysit.
Frequently asked questions
What is a software factory for agents?
It's a pipeline that turns an idea into shipped code with AI agents doing the building: you describe the work, it's broken into bounded tasks, agents build each one in an isolated sandbox, and only changes that pass your tests and an independent review get merged. partyline is that pipeline, from describe to pull request.
How is this different from just running an AI coding agent?
A single agent in a terminal builds one thing at a time and asks you to trust its output. The factory adds the parts a team needs: an interview that turns a vague idea into a real plan, isolated worktrees so many agents build in parallel without colliding, a verify gate that runs your tests plus an independent reviewer, and a quarantine that holds anything that fails instead of shipping it.
Where do the agents run, and what does partyline see?
Agents run on your hardware, in your git, with your keys. The server coordinates the work — the plan, the queue, the run log — but never holds a model key. The only key it holds is the GitHub App private key used to open pull requests.
Which AI agents can I use?
Claude Code, Codex, and Gemini run as build agents today, plus open-weight models and Grok through the llm bridge. Shared-context threads currently run with Claude Code (Codex is experimental) and are built on open MCP, so more engines read the same context as their support lands.
What stops a wrong change from getting merged?
Every task's output runs two gates: your test suite and an independent adversarial reviewer that tries to find what's wrong. A change only becomes a mergeable PR if both pass. Anything that fails is quarantined and routed to a human to approve or discard — the pipeline never merges on trust.
Do I have to use the whole pipeline?
No. You can describe and plan without building, run a single task, or use just the session manager, shared context, or terminal sharing on their own. The factory is what you get when you connect them end to end.
Describe the work. Ship what passes.
$ brew install partyline-sh/tap/partyline