stui

The open-source, single-machine edition (Apache-2.0). Your tmux sessions and the coding agents in them, their orchestration runs and CI, with no account and no relay.

stui is a terminal UI for tmux and the coding agents running in it. It shows your sessions, windows and panes, which Claude Code, Codex, opencode or Pi agents are idle, working or asking for you, the GitHub Actions state of each pane's repository, and the orchestration runs where one agent hands work to others. It is the single-machine edition: no st, no account and no relay. It is open source, Apache-2.0, at sophonz-labs/stui. The current release is v0.2.3. How it compares with tmui and st ui is in the table on the tmui page.

Everything stays on this machine. stui runs a small server on 127.0.0.1 that watches tmux and the agents' hooks, and keeps its state in ~/.stui. It makes two kinds of network call: gh asks GitHub for CI runs, and the usage poller asks Anthropic and OpenAI for your plan limits (stui usage poll off stops it).

Install

curl -fsSL https://dl.sophonz.io/stui/install.sh | sh
stui hooks install    # let Claude Code, Codex, opencode, Pi and others report their turns

The script puts stui in ~/.local/bin after checking its SHA-256 (STUI_INSTALL_DIR, STUI_VERSION=v0.2.3). With Go: go install github.com/sophonz-labs/stui@latest. Without hooks stui still reads the agents' screens, but hooks are what tell working from asking reliably.

stui upgrade replaces it with the latest release the same way, checking the SHA-256, and restarts the background server on it. A stui that Homebrew installed is left alone: it says to run brew upgrade stui.

It needs tmux 3.0 or later (3.2 to attach in a popup) on macOS or Linux, and gh for CI.

Using it

stui                  # the UI; starts the background server if needed
stui ui agents        # open on a view: sessions, agents, orch, ci, usage
stui status           # the server and every agent's hooks
stui stop             # stop the server

1–5 switch views: sessions, agents, orch, ci, usage. enter goes down a level and esc comes back. ? lists every key. working is teal, asking (the agent waits for you) is orange. A state the pane's program reported itself (OSC 7501) is marked, working (reported) (v0.2.3). Claude Code does not report this way inside tmux, so stui reads it from its hooks and screen.

stui's agents view: the header with the OS, tmux, the panes, CI and the Claude plan limits, the agents' counts and the keys; below, five agents in the session work with their window, agent, state (one claude asking, three working, one idle), turn time, CI state, orchestration role and title — The agents, their CI and their plan limits

Key	Where	Does
`a`	sessions, windows, panes, agents, orch	attach; `prefix d` comes back, as in tmui
`c`, `o`	panes, agents, CI	that pane's CI runs; open the run in the browser
`n`	sessions, windows	new session; new window, optionally with an agent
`ctrl-d`	sessions, windows, orch	kill or stop (asks first)
`r` `d`	orch	read a worker's answer, its diff
`ctrl-e`	everywhere	the header on and off
`:`	everywhere	a view, `:ext`, `:config`, or any stui command: `:worker list`, `:orch history`

Copying is the same as in tmui: y and Y copy a row, v and y select and copy a pane's output, and z shows the output alone for your terminal's own selection.

stui showing an agent's pane: Claude Code has updated two files and waits on a permission prompt to run npm run build — An agent's pane, waiting on a permission prompt

Orchestration

One agent, the coordinator, starts other agents as workers in their own tmux windows, hands them tasks, waits, and reads their reports. The commands are for the agents; stui plugin install teaches Claude Code and Codex to use them.

stui orch start --objective "make the parser tests pass"
stui worker start --agent claude --name tests --task task.md
stui worker wait --check-in 3m
stui worker read tests

stui's orchestration view: a run with a coordinator and four workers api, web, docs and tests, each with its pane, agent, state, turn time, window and summary — An orchestration run: a coordinator and its workers

With st on the same machine

When st runs on the same machine, stui shows st's runs and agents too, read-only. It asks st's local server (on 127.0.0.1, GET only) at every refresh, and marks st's runs (st).

Reading works on st's rows: output, attach, CI, copy, and r and d.
Changing does not: ctrl-d on one of st's workers says this is st's run: use st ui. A run belongs to the tool whose coordinator started it.
st.read: false in the config turns it off.

Other machines stay st ui's: stui is this machine only.

Plugins and config

stui reads the same plugins as tmui (tui-plugins): stui ext install|list|remove <name>. (stui plugin is something else: the orchestration skills for the coding agents.)

Settings live in $XDG_CONFIG_HOME/stui/config.yaml (~/.config/stui/config.yaml by default), with stui config and stui config edit. The keys are tmui's — ui.logo, ui.headless, ui.color, refreshRate, startView, plugins.enabled, plugins.disabled — plus st.read. startView takes sessions, agents, orch, ci or usage.