You are the stakeholder. Chillax runs the company.
A terminal manager for a team of AI coding agents. You hire a project manager, developers, a tech lead and QA. They plan, build, review, test and release. You approve the sprint and the release.
Source is private while rough edges get fixed. Star or watch the repo to hear when it opens.
A software company where every employee is a coding agent
Chillax doesn't rebuild a coding agent. It manages the ones you already use. Each hire is an existing agent, driven over the Agent Client Protocol, with its own role, policy and model.
claude-codecodexcommand-codeclaude-codeclaude-codeAn example team. Each hire can use any supported agent and model.
| In a company | In Chillax |
|---|---|
| Job description | A role file: a system prompt plus a policy |
| Employee | A hired coding agent: Claude Code, Codex, Command Code and others |
| Jira | A built-in local tracker with epics, stories, bugs and sprints |
| Dev, QA, prod | Git branches, worktrees and your project's own commands |
| Company handbook | Chillax's MCP tools, which agents use to follow the process |
Two decisions are yours. The rest runs on its own.
Everything between the gates is the team's job. Nothing starts until you approve the sprint, and nothing ships until you approve the release. Try them.
- You describe what you wanta chat with the PM, or a plan you already wrote
- The PM plansepics, stories, acceptance criteria and a proposed sprint
-
Sprint gateWaiting on you: 6 tickets proposed for SP-1
- Developers buildone
git worktreeand branch per ticket, your tests run - The tech lead reviewsapproved diffs merge into
dev, your checkout untouched - QA testspromotes
devtoqa, checks acceptance criteria, files bugs back -
Release gateWaiting on you: v0.1.0 passed QA
- Shipped
qamerges tomain, tagged, with release notes
What you get
- Hire anyone
- Pick the agent and the model per hire. Cheap models for simple tickets, strong ones for review.
- An inbox for you
- Permission requests, questions and both approvals wait in one place, so you only look when you're needed.
- Watch it live
- The same views in the terminal and the browser: board, team, chat, a live feed of every agent's tool calls, and token usage per ticket.
- A bug loop
- QA files bugs as tickets and they go back to a developer, like they would in a real team.
- Pause, quit, resume
- Every change is an event in a local log. Quit any time and the team, chats, tickets and inbox come back where you left them.
- Ask for changes mid-sprint
- New requests become tickets. If an agent gets stuck, it can escalate or hire help.
The full walkthrough
Five minutes, from a written plan to a tagged release: planning, hiring, both gates, the dev loop, the live view, QA and a finished project.
The honest trade-off
Chillax plans heavily, so it uses more tokens and more time than handing one prompt to one agent. In return, every piece of work has a ticket, the reasoning lives in its comments, and nothing reaches main without you.
Where it is now
Chillax is an MVP. Planning, hiring, both gates, the dev loop, QA and the bug loop work and are covered by end-to-end tests. Real agent runs have taken a project from the PM's plan to a tagged release. Next: Jira and GitHub tracker adapters, more agent backends, budgets and crash recovery.