swe-mux

A coding-agent multiplexer built for desktop and mobile.

Run Claude Code, Codex, opencode and any other CLI side by side in real terminals, see which one needs you, and drive the whole thing from your phone. It runs on your own machine: no vendor backend, no relay, no account, no telemetry.

_{A silent seventy-second walkthrough - play it on swemux.dev. GitHub only plays video uploaded through its own web UI, so this is a still that links to the player rather than a video tag that would render as a dead frame here.}

Why it is different

• Real terminals, not a re-render. swe-mux owns the pseudoterminals, so your CLIs behave exactly as they do outside it - including whatever shipped yesterday. There is no feature lag and nothing to wait for us to expose. Start a plain shell and promote it to an agent mid-session, because the terminal is the primitive.
• One status vocabulary across every vendor. Working, ready, awaiting you, or blocked - read from provider hooks, the transcript, the terminal and the CLI's own state, with every transition in a durable ledger. Ambiguous evidence resolves to the conservative prior rather than to a guess. (how)
• Built for mobile, not shrunk onto it. An installable PWA over your own Tailscale tailnet with nothing turned off: terminals, git review, the editor, previews, voice, push. A dev server on 127.0.0.1 is proxied through swe-mux's own URL, so your phone reaches it without exposing a port. (how)
• Sessions that outlive the app. A supervisor process separate from the daemon holds every pseudoterminal, so a daemon restart or a full rebuild leaves the agents working - new builds of swe-mux ship from an agent session running inside swe-mux. Reconnecting replays only the bytes you missed. (how)
• Your workflows stay yours. Harness-neutral by construction: normalized input, status, transcripts, history and accounts across the CLIs in its registry, and anything else runs unchanged. Switch harness or model without rebuilding how you work. (how)
• Parallel worktrees, landed behind your gate. One branch at a time: reconcile, run the verification command whose exact bytes you approved, fast-forward only. A conflict or a failed gate goes back to the branch's own agent, and an agent cannot approve the gate its own land runs. (how)

Underneath those, a record taken from the work rather than from the agent's account of it: every file write hashed on the bytes actually written, every command with its exit class, test output parsed to the failing set, and each commit attributed to the session and conversation that produced it. (Tier 0 facts)

The full feature list is at https://swemux.dev; this file is the short version.

Almost everything beyond the workbench is off until you ask

Worth knowing before you install rather than after. Automations are per-Project opt-in and every one ships off. The model-backed capabilities - the behaviour timeline, the attention observers, and the assistant - ship off. Read aloud ships off. The land queue needs four separate things before an agent can trigger one.

Nothing runs on a Project that did not opt in, and nothing reaches a model without a budget you set.

What crosses the network: your data is SQLite on your disk, there is no swe-mux account, and nothing reports usage anywhere. Your CLIs talk to their own vendors under your own subscription. OpenRouter, web push, Hugging Face model downloads and Edge TTS are each optional and off until you turn them on. The one request swe-mux makes for itself is a daily static version.json fetch carrying no identifier, and it is disableable.

Install

swe-mux is on PyPI. The wheel is pure Python and carries the built frontend, so this needs no Node and no checkout.

# Recommended. Isolated environment, every command on your PATH globally.
uv tool install swe-mux

# The same isolated, on-PATH install, without uv.
pipx install swe-mux

On Windows that is everything: run swe-mux and you get the native window and a tray icon, with no console and nothing else to install. It offers to add itself to the Start Menu and to start with Windows the first time, so after that there is nothing to launch at all.

Elsewhere - or if you would rather just use the browser - swemux start puts the daemon in the background and returns once it is serving http://127.0.0.1:8765; closing the terminal does not stop it, and swemuxd --shutdown does. swemuxd still runs it in the foreground when you want the log in front of you. swemux doctor is a read-only health report covering the daemon, the supervisor, the frontend build, detected agent CLIs, the tailnet listener, and background loops.

Every install writes three commands, one per program: swemux (the CLI), swemuxd (the daemon), and swe-mux (the desktop window and tray). There are no short aliases, deliberately: mux is shared with at least one unrelated tool, and shipping a launcher under a contested name is what creates the collision rather than what survives it.

A Windows installer is published from v0.1.2 onward, alongside a portable archive, on the releases page. It is not code signed, so SmartScreen warns on first run; the PyPI install avoids that prompt entirely.

No Python install of any kind creates a desktop shortcut. Wheels have no post-install hook, so that is structural rather than an oversight: start swe-mux from a terminal, or run swemux install-shortcut.

Running from a checkout, upgrades, PATH troubleshooting and uninstall: .docs/development/OPERATOR_LIFECYCLE.md.

Requirements

• Python 3.12+, and uv or pipx to install with.
• At least one agent CLI, already installed and logged in. swe-mux does not install, manage, or proxy them.
• Optionally Tailscale, to reach the daemon from a phone.
• Node 22.6+ only to build the frontend from source; the published wheel carries it.

On-device speech is optional and acquired on an explicit press rather than shipped: the libraries and their models are downloaded once, verified against pinned hashes, and nothing is fetched until you ask. Speech-to-text decodes on your own machine in both shipped configurations - faster-whisper, or Windows Speech Recognition. There is no cloud speech path and no browser fallback; without an engine, transcription returns a typed error rather than sending audio anywhere. Text-to-speech is the same, except the explicitly experimental Edge TTS provider, which does leave the machine and requires an acknowledgement first.

First run

Create a Project and point it at an existing folder. Ctrl+Alt+T opens a terminal at its root, Ctrl+Alt+P opens the command palette; nothing is spawned until you ask.

Then type claude, codex, or another CLI normally. swe-mux puts its own launchers first on that terminal's PATH, so the usual command promotes the terminal you are standing in to an agent session in place: same pane, same scrollback, now carrying a transcript, a status, a queue, and a context meter.

The Run menu starts an agent, a shell, a worktree session, or an imported task. Imported tasks - VS Code tasks, package.json scripts, .swe-mux/actions.toml - stay inert until their exact current bytes are trusted. (project actions)

Platform support

CI builds, validates and install-smokes the wheel on Windows, Linux and macOS every push, and starts a real daemon on Linux and Windows from the source checkout to prove it serves a shell and exits cleanly. No CI job starts a daemon from a published artifact, which is exactly where the proof stops.

• Windows 10/11 is the proving platform. The full gate runs there, including real ConPTY integration and the Playwright renderer suite, and it is the only platform the desktop app ships on.
• Linux runs headless plus a browser, on a required CI leg. There is no Linux desktop app, by design.
• macOS is implemented, typechecked and exercised, but its leg is still continue-on-error; treat it as unproven.

What each claim rests on: .docs/development/CROSS_PLATFORM_FINDINGS.md.

Remote access

The daemon listens on localhost and on the machine's detected Tailscale IPv4 address; swemuxd --local-only keeps it local. Tailscale policy is the access boundary, and swe-mux has no separate remote login, so a tailnet peer your policy admits has terminal and code-execution authority.

Browsers restrict the clipboard and microphone over plain HTTP, so for those put Tailscale Serve in front: tailscale serve --bg http://127.0.0.1:8765. 0.0.0.0, direct LAN binding, Funnel, port forwarding and public ingress are unsupported. (detail)

Configuration and data

Configuration is config.toml inside the data directory: ~/.mux on Windows, $XDG_DATA_HOME/swe-mux (else ~/.local/share/swe-mux) on Linux, ~/Library/Application Support/swe-mux on macOS. An existing ~/.mux always wins, and MUX_DATA_DIR overrides all of it.

Provider accounts

Settings → Accounts saves Claude and Codex logins, tracks each account's subscription window, and switches the system-wide login in one click. Only authentication is copied.

This is a convenience for one person switching between accounts they personally own and pay for, replacing the logout/login cycle the provider CLIs otherwise require. It is not account pooling and not a way around a usage limit: accounts are never shared between people, credentials stay local and go nowhere but the provider's own endpoints, sessions are never load-balanced, and switching is always explicit. (scope and terms)

Documentation

Published docs: https://swemux.dev/docs/. The maintained design contract starts at .docs/design/00_OVERVIEW.md, and .docs/CLAUDE.md routes each subsystem to the document that owns it.

Contributing

Contributions are welcome. Feature requests are Discussions in the Ideas category, where a thumbs-up is a vote and the most-voted open ideas are drawn on the roadmap; issues are for bugs. Describe the problem rather than the fix. CONTRIBUTING.md covers what a change has to satisfy; CLAUDE.md covers the working rules; SECURITY.md is where a vulnerability report goes.

License

swe-mux is licensed under the Apache License 2.0. See NOTICE for attribution and TRADEMARK.md for what the license's trademark reservation does and does not allow.

Contributions arrive under a DCO sign-off - git commit -s - and not a CLA. A CLA would let the project relicense your contribution later; a DCO does not, and that is the intended trade.

Third-party software redistributed with swe-mux is listed in THIRD-PARTY-NOTICES.md, generated from the lockfiles by packaging/license_audit.py so it cannot drift. swe-mux ships no GPL or AGPL code; it ships pystray under the LGPL as replaceable source inside the bundle.

Not affiliated with the agent vendors

swe-mux launches and observes coding-agent CLIs published by other vendors, including Anthropic's Claude Code and OpenAI's Codex CLI. It is not affiliated with, endorsed by, sponsored by, or certified by Anthropic, OpenAI, or any other such vendor, and it uses their names only to identify which tool a feature works with. You run those CLIs under your own account and your own agreement with each vendor, and the same is true of the optional OpenRouter and Hugging Face integrations.

The optional Edge TTS integration is different: the upstream client uses Microsoft Edge's consumer Read Aloud endpoint with no API key and no documented third-party service contract, so selecting it requires an explicit disclosure acknowledgement. That client is LGPL, runs only in an isolated managed or operator-supplied Python, and is absent from the frozen bundle - but that software boundary does not resolve Microsoft's service terms.