# rlx: a manager for your AI coding agents

> rlx is a manager for your AI coding agents. Tell it the goal: it runs Claude Code and Codex on accounts with room. Then relax.

- **URL:** https://aisquare.studio/rlx
- **Audience:** Developers who run coding agents such as Claude Code or Codex and want a manager to run them
- **Topics:** rlx, RLX, relax, relax AI, relax AI agents, AI agent manager, AI coding agents, coding agent orchestration, multi-agent, Claude Code, Codex, usage limits, account headroom, coding agents, agent manager, parallel agents, account handoff, headroom, git worktrees, task board, terminal, tmux, open source, MIT

## Tell the manager the vision. Then relax.

More than 10x the agent work (about 12x in our internal dev teams' testing, per week). A third of the babysitting. The source, in a card under ">10x" (hover, focus or tap): From our internal dev teams' testing: by hand Jul 4 to Sep 13, with rlx Oct 12 to Oct 30, 2026. Claude tokens a week: 2.9B by hand, 34.6B with rlx, about 12x. Where the fleet's tokens went: 39% checking, 25% managing, 36% building. Sessions you actively managed a week: 46 by hand, 17 with rlx.

Plans the work. Spawns Claude Code and Codex. Moves an agent at its limit to an account with room. You approve the merge.

Install, two options. macOS / Linux: `curl -fsSL https://aisquare.studio/rlx/install.sh | sh` (installs only what's missing; safe to re-run; works on WSL2). Windows, in PowerShell: `irm https://aisquare.studio/rlx/install.ps1 | iex` (runs the installer in WSL2, or prints how to get WSL2). The header's Install button copies the command into the install block at the top of the page (no scrolling when it is in view; from lower down it jumps up to it) and opens the visitor's tab there.

MIT · one SQLite file on your machine · [read install.sh](https://github.com/AISquare-Studio/rlx/blob/main/install.sh).

CTAs: [Star on GitHub](https://github.com/AISquare-Studio/rlx), Watch it run (the demo, on this page) and [Docs](https://docs.aisquare.studio/docs/rlx/overview/).

On desktop, beside it, rlx's real screens over made-up data, looping from Welcome to READY; on smaller screens and without JavaScript, a sample run, as the board feed prints it on a made-up clock: 09:02 you ask to add passkey sign-in to the web app; 09:03 the manager puts 4 tasks on the board, each with a contract; 09:04 it spawns coder-3 on acct 2, the first under 85%; 09:11 coder-3 claims T-7; 10:26 coder-1's acct 1 hits its limit and it moves to acct 2 and resumes; 11:32 the reviewer approves T-7 with CI green; 11:47 READY, 2 PRs with the evidence, your call.

Keyboard: ? opens the shortcuts, Cmd/Ctrl+K jumps to any section or copies the install (first item: your system's).

## Wins

- **Limit hit? Handed off.** An agent at its usage limit moves to an account with room.

- **A worktree each.** Each coder and reviewer gets its own git worktree.

- **No merge without your ✓ and green CI.** It merges only after a person approved it and CI is green.

- **Click in. It's real Claude Code.** Click any agent to type in its real Claude Code or Codex session.

- **Fleet headroom, in your status line.** The opt-in Claude Code status line: `acct 3 · 5h 42% · wk 61% · fleet 58% left (3×) · ctx 34%`. Turn it on with `rlx accounts statusline --install`.

Also: Steer it from your phone, Pause all, resume all, Memory across sessions, Local-first, MIT.

## Limit hit? Already handed off.

Watch one handoff. Then hold 🔥 and burn the rest. A toy with three made-up accounts: coder-1 and coder-2 on acct 1, coder-3 on acct 2, none on acct 3 (Codex). Each account burns in step with the coders on it, so two coders burn twice as fast as one. On first view it plays one handoff: acct 1 burns to its limit and its coders resume their sessions on acct 2, which then burns faster. Then hold to burn: with every Claude Code account full, they start fresh on Codex with a brief. Readouts: session kept, handoffs.

An agent at its usage limit moves to an account with room and resumes its session, unless the limit lifts within 15 minutes.

## Relax: 4 terminals → 1 manager.

You're in it twice: the goal, and the merge.

Before and after, side by side behind a drag divider, each named above it. Before (without rlx): coder-1: usage limit, waiting for you to switch accounts; coder-2: waiting for your permission; coder-3: lost its task, waiting for your input; you, copying between them (⌘C ×14), 3 waiting. Each terminal's border and bell blink for attention.

After (with rlx), calm: manager: 4 tasks, 3 agents; coder-1: limit hit, resumed on acct 2; reviewer: approved, CI green; you: ☕; 0 waiting.

Each coder and reviewer gets its own git worktree, so two coders never edit the same files. The manager never writes code or pushes. It merges a PR only when a person approved it on GitHub and every CI check on its head is green.

## Watch it run.

One goal, from Welcome to READY, in rlx's real screens over made-up data. In the app, an agent's pane is its live session. [The fleet view in the docs](https://docs.aisquare.studio/docs/rlx/fleet-ui/).

A scroll story: the screen stays in place while the section scrolls by, and the scroll alone moves the story a step forward or back. The camera zooms and tilts into the part of the screen each step is about; at step 6 it dives into the sidebar's accounts and the Accounts page takes over. Nothing in the screen is clickable. Beside it (under it on a phone), the step's caption and a rail of the nine steps, each a link to its step. With reduced motion, or without JavaScript, the steps are a plain list of the screens with their captions.

- 1 · Welcome: the project, the coding agent, the fleet, one step each.

- 2 · The goal: you give the manager the goal, in plain words.

- 3 · The board fills: the manager plans: tasks on the board, each with a contract.

- 4 · Coders claim: coders claim the tasks; the live feed shows every claim.

- 5 · A coder at work: inside coder-auth's real Claude Code session, writing and testing.

- 6 · A usage limit: coder-auth hits its account's usage limit.

- 6 · The accounts: personal used up; 42% of the fleet left.

- 7 · Moved, resumed: moved to an account with room. Same session, back at work.

- 8 · Review: tester-auth's own session checks the work against its acceptance.

- 9 · READY: the PR and the evidence. No merge without your approval.

- 9 · The manager: the team has exited; it waits on you.

- 9 · READY: the manager's note names PR #42 and PR #43, all done and tested.

## Everything is a command.

A playable terminal. It opens on `rlx accounts usage`; `rlx --help` prints rlx's own help, laid out at the terminal's width (every command takes the global --json); the rest is made-up output: type or click a command, Tab completes, Up recalls; --json shows the same as JSON. A few in-character answers too: `git push --force` gets "the manager never pushes. it waits for your ✓.", `rlx relax` gets "☕ the manager has it."

- `rlx fleet ls`: this project's agents, live

- `rlx board`: who took what, where it stands

- `rlx accounts usage`: five-hour and weekly use

- `rlx fleet top`: CPU, memory and tokens per agent

- `rlx doctor`: every check, with its fix

- `rlx outcomes`: this week's work, and who steered

- `rlx remember --user "prefer pytest"`: a preference every session starts with

## One line. You're in.

Paste it, then type rlx. It installs only what's missing (uv, its own Python, rlx, tmux, gh, git, Claude Code, Node (optional)), then offers to open rlx. A sample run: Checking this machine can be installed onto / ✓ curl present / Detected macos/arm64 / Surveying what is already installed / ✓ git · ✓ gh · ✓ Claude Code / + uv · + Python · + tmux · + rlx / Open rlx now? [Y/n] y.

Welcome walks you through project, agent and fleet.

## FAQ.

- **What is rlx?** A manager for your AI coding agents, Claude Code and Codex, in one terminal.

- **Which AI coding agents does rlx run?** Claude Code and Codex. Click any agent to type in its real session.

- **What happens when an agent hits its usage limit?** It resumes on an account with room, unless the limit lifts within 15 minutes.

- **Does the manager write or merge code?** Never writes code or pushes. Merges only after your approval on GitHub and green CI.

- **How do I install rlx?** macOS / Linux: curl -fsSL https://aisquare.studio/rlx/install.sh | sh. Windows, in PowerShell: irm https://aisquare.studio/rlx/install.ps1 | iex (it runs the installer in WSL2).

- **Where does rlx keep its data?** One SQLite file on your machine. No daemon, no cloud.

- **Is rlx open source?** Yes, MIT. The source is on GitHub: AISquare-Studio/rlx.

## go relax.

The manager runs the work. No merge without your ✓ and green CI. Install, Star on GitHub, Share.

## Links

- [rlx docs](https://docs.aisquare.studio/docs/rlx/overview/)
- [Install](https://docs.aisquare.studio/docs/rlx/install/)
- [Quickstart](https://docs.aisquare.studio/docs/rlx/quickstart/)
- [Accounts](https://docs.aisquare.studio/docs/rlx/accounts/)
- [Command reference](https://docs.aisquare.studio/docs/rlx/reference/)
- [Source on GitHub (MIT)](https://github.com/AISquare-Studio/rlx)

---

**Machine-readable AISquare**

- Site index for LLMs: https://aisquare.studio/llms.txt
- Agent instructions (when to use AISquare): https://aisquare.studio/agents.md
- OpenAPI specification: https://aisquare.studio/openapi.json
- MCP endpoint (Streamable HTTP, POST): https://aisquare.studio/mcp
- Sitemap: https://aisquare.studio/sitemap.xml
- Blog feed (RSS, full posts): https://aisquare.studio/blogs/rss.xml
- Any page on this site also serves markdown: append `.md` to its path (the homepage is `/index.md`) or send `Accept: text/markdown`.
