ACTower supervises your coding agents, and gives them and your team a shared wiki. You run AI coding agents in terminal panes; ACTower watches them, answers the safe requests itself and brings the rest to you in one place. It runs on your own machine.

What ACTower does (2 min 4 s). Real ACTower screens, sample data. Recorded 2026-11-12.

The hard part is supervising them

Running more coding agents isn't the hard part anymore. Supervising them is: every new agent is one more terminal to watch.

  • Which agent needs me? A question waits in a pane you aren't looking at.
  • Which one is done? An agent finishes while you're watching another window.
  • What did each one do? At the end of the day you can't say what each agent asked to do and how it was answered.

ACTower answers all three on one screen.

Most of it never reaches you

  • Each agent CLI gets its own classifier, built for the way that CLI asks. When an agent asks to read a file, run a command or edit code, ACTower reads the exact request your agent drew, checks every part of a command, and approves only what it can vouch for. Anything risky, or anything it isn't sure of, comes to you.
  • The safe ones are approved on the spot. Reading files and folders, searching, fetching a web page, and shell commands that only read: the agent keeps working, and you never see the prompt. Edits and writes wait for you unless one of your own rules covers them.
  • The risky ones wait for you, in one queue, with the pane's context beside them. Some actions, such as deleting files, sudo, a force push or network commands, can't be approved by any rule you add.
  • The time adds up. In our own use since July, with most of our work in Autopilot sessions, ACTower answered about 80% of agent requests without a person: more than 150,000 of them. By Activity's estimate, that saved about 870 hours. Activity shows your own counts.
  • You decide where it answers. Auto-respond is on from the first start. Turn it off, or choose which sessions it answers, in Settings → Responder. By default it follows the sessions the Monitor shows.

Multiplexers help you see multiple agents. ACTower helps you avoid watching all of them.

Monitor: every agent at a glance

  • One card per pane, with its state: Working, Question or Idle. The Recently Done chip shows the agents that finished in the last 10 minutes.
  • Your agents, your terminals. Claude Code is fully supported; Codex, Google Antigravity CLI and Cursor are in beta. ACTower finds them in tmux and in Herdr (0.9.0 or newer).
  • Three views. See the panes by session, by repo, or as your window layout.
  • Go to pane. The terminal icon on a card takes you to that pane: ACTower brings its terminal window to the front, or opens one if none shows it. In Herdr it focuses the pane in Herdr's own window.
  • Done, out loud. When an agent finishes, a sound and a toast tell you which one, and the toast has its own Go to pane.

Responder: one queue for every question

  • Every question that needs you lands here, from every pane, with a preview of the pane so you see what the agent sees.
  • Answer without hunting for the window. Approve or Deny, or type an answer when the agent asks for one. Skip presses Escape in the pane: the agent drops that request and waits for your next message.
  • Check what was answered for you. The Recently Auto-Approved list shows each request ACTower approved on its own.

Activity: what each agent asked, and how it was answered

  • A summary for today or the last 7 days: total actions, how many were answered automatically (and the share inside Autopilot), clicked by a person in ACTower, or denied and skipped (the rest you answered in the terminal yourself), with an estimate of the time saved. Below it, Activity shows what your agents asked, by type and by risk, with actions per day and the busiest agents.
  • The decision log records every request with its own ID (it's in the JSON export), per pane: who answered it (ACTower's rules, Autopilot, you in ACTower, or you in the terminal) and how. Turn on Save decision captures in Settings and each entry gets View moment: the pane as it looked when the question came in.
  • Local, and yours. The logs stay on your machine, and you can export them as JSON.

Custom Safe Rules (Plus)

Teach ACTower the commands that are safe in your own projects, and it approves them for you too. Your rules can't approve the dangerous ones: deletes, sudo, force pushes, network commands and the like.

Beyond supervision

  • Hand Autopilot (Premium) a whole task: a developer agent builds it in its own copy of the repo, reviewer agents check it, and you stage the result and commit. See Autopilot.
  • Ecosystems give your agents and your people a shared wiki that grows with every task. See Ecosystems.

Where it runs

ACTower runs on your machine, as the desktop app and a local web app, with terminal commands for the main jobs: monitor, respond, report, rules and Autopilot. The Free plan watches up to 5 agents and includes auto-approve, local logs and Ecosystems. Plus adds unlimited agents and Custom Safe Rules; Premium adds Autopilot. See License & Billing.

ACTower is the supervision and workspace layer for teams of coding agents and humans. It supervises agents while they work, can run reviewed autonomous work through Autopilot, and gives agents and humans a shared organizational memory through Ecosystems.

Next: Getting Started.