MuxAgent is a CLI for running coding agents through graph-based workflows. Use graph-based workflows to plan, review, approve, implement, and verify code with Codex or Claude Code.
- Task System — Run code tasks through graph-based workflows with explicit planning, review, approval, implementation, and verification steps. Three ready-made configs cover different risk tolerances. Supports Codex and Claude Code runtimes.
- Remote Control — Monitor and control Claude Code sessions from your phone via a paired mobile app.
curl -fsSL https://raw.githubusercontent.com/LaLanMo/muxagent-cli/main/install.sh | shThe install script puts muxagent in /usr/local/bin when writable, otherwise
it falls back to ~/.local/bin.
Download the latest muxagent-windows-*.zip bundle from
GitHub Releases, unzip it,
and run muxagent.exe.
Official installs include everything needed to run MuxAgent with Claude Code.
muxagentThis opens the workflow CLI. Pick a task config (default, autonomous, or
plan-only), describe the task, and MuxAgent routes the agent through the
workflow for you.
-
Download the MuxAgent mobile app. Google Play | iOS coming soon.
-
Run:
muxagent daemon start
-
Scan the QR code in the app to finish setup.
On a new machine, muxagent daemon start begins first-time setup, shows a QR
code, waits for approval in the mobile app, and then starts the daemon.
You can also run muxagent auth login manually if you want to pair before
starting the daemon.
A task config defines a workflow graph — the sequence of nodes and the edges between them that an AI agent follows. MuxAgent ships four ready-made configs:
default — When you want human sign-off before code changes land.
┌─────────────────────────┐
│ (approval rejected) │
▼ │
plan ──▶ review ──▶ approve ──▶ implement ──▶ verify ──▶ done
▲ │ ▲ │
└─────────┘ └──────────────┘
(review rejected) (verify failed)
autonomous — When you trust the agent and want fast iteration.
plan ──▶ review ──▶ implement ──▶ verify ──▶ done
▲ │ ▲ │
└─────────┘ └──────────────┘
(review rejected) (verify failed)
plan-only — When you want a reviewed plan without touching code.
plan ──▶ review ──▶ done
▲ │
└─────────┘
(review rejected)
yolo — Fully autonomous multi-wave mode. No approval, no clarification.
┌──────────────────────────────────────────────────┐
│ (next wave) │
▼ │
plan ──▶ review ──▶ implement ──▶ verify ──▶ evaluate ──▶ done
▲ │ ▲ │
└─────────┘ └──────────────┘
(review rejected) (verify failed)
Workflow configs are different from runtime selection:
- a workflow config chooses the graph, bundled prompts, and product intent
- runtime selection chooses which coding runtime executes agent nodes, for example
codexorclaude-code
The included workflow configs are stored as task config bundles under ~/.muxagent/taskconfigs.
You can clone them and modify the YAML to change the workflow graph, prompts,
runtime, iteration limits, or clarification settings.
If you already have a user config named plan-only, autonomous, or yolo, MuxAgent
preserves it and installs the built-in config under a fallback alias such as
builtin-plan-only. Existing bundle files are never overwritten.
See Task Config Semantics for the full edge, iteration, and schema specification.
Task TUI
muxagent— Launch the interactive workflow CLI.
Daemon
muxagent daemon start— Start first-time setup or start the daemon.muxagent daemon status— Show daemon status.muxagent daemon stop— Stop the daemon.
Auth
muxagent auth status— Show pairing status.
General
muxagent version— Show the installed CLI version.muxagent update— Updatemuxagent.