Skip to content
encs – Přepnout do češtiny
cs – Přepnout do češtiny
back to homepage

Watch every Claude Code session, answer its approvals right from the menu bar.

View on GitHubDownload for macOS
$ brew install --cask michalstrnadel/tap/agentbar

Free & open source · macOS 12+ · No telemetry · v1.56.0

One queue, under the notch. Allow in one click, or deny and say why. A rule that only watches keeps a record instead.
Works with
  • Claude Code
  • Codex
  • Cursor
  • Gemini
  • Copilot
  • OpenCode
  • Qwen Code
  • Devin

Say yes without leaving what you're doing.

The request comes to you, with the exact command. One click, no hunting for the right terminal tab.

Deny with a note Type “use pnpm here, not npm”. The agent reads it as the reason and changes course.
Hand it a file Drop a screenshot onto a session. Its path lands in that agent's prompt, never with a Return.
Or from your phone Off until you switch it on, and only while you're away. Allow comes only with a command short enough to read whole.

Rules you wrote. Nothing else.

AgentBar never answers on its own. A rule you typed may, and every time it does, it writes down which one.

Try a rule before it speaks Type a command into the rule and see on the spot whether it would have taken it, and why.
Then let it watch It answers nothing and keeps score: would have allowed 14×, you did the same 14×. Let it answer when that holds.
  • Watching first

    A new rule only notes what it would have done. You let it speak when a week of that looks right.

  • The live command

    Before anything is allowed, the real command is checked again, not just its shape.

  • On the record

    Every firing is a row in the ledger, naming the rule that fired.

Your Mac stays awake while your agents work.

One click on the cup in the island, or let it start by itself when an agent works. While you're away the keys go dark. If you want, it puts the Mac to sleep five minutes after the last agent finishes, or locks the screen when you leave.

While agents work Or for 15 minutes, an hour, until 18:00, or until you say. It can also start on the charger, with a display, or while an app runs.
With the lid closed Close the MacBook and the agents keep working. You enter your password once to turn it on, never to turn it off. It ends with the mode, after 12 hours, or if the Mac runs hot.

Meet Clawd, who lives under your notch.

Reading, editing, delegating, waiting on you: you can tell what an agent is doing from across the room.

Scenes Every state has its own little scene, drawn pixel by pixel.
Twelve states Thinking, reading, searching, editing, running a command, browsing, delegating, compacting, using a tool, needs approval, has a question, asleep.

Waiting on a build? Take a break.

Two tiny games inside the island. They pause when you look away and step aside the moment an agent needs you.

Space Bugs Clawd vs. the bugs. The joystick beside ⋯ starts it; arrows to move, Space to fire, and the high score stays on your Mac. The bugs are, of course, your bugs.
Bug Hunt Ten bugs a round, three shots at each. Let one fly off and the dog laughs.

Your day with agents, on one card.

Agent time, the hours, your top agent, what changed, how you answered. Copy it, post it, or save it as a video.

Your Day Or your week. Opened from the menu, never on a schedule.
A week of decisions: allowed, denied and answered by a rule, per day.
A week of decisions What you allowed, what you denied, what a rule answered for you. Per repo, per day.

Fair questions.

The ones people ask first, answered the way the app behaves.

Is AgentBar free?

Yes. It is free and open source under the MIT license, with no account and no paid tier.

Which AI coding agents does it work with?

Claude Code and Claude Cowork, Codex, Cursor CLI, Gemini CLI, GitHub Copilot, Google Antigravity, Qwen Code and OpenCode, plus Cursor cloud agents, Devin and Codex cloud tasks. Any other tool can join with one line of the agentbar report command.

Can I approve every agent's requests from AgentBar?

Claude Code, Codex and GitHub Copilot can be allowed or denied in a click. Cursor's and Gemini's hooks can only refuse, so for them AgentBar shows the status and the question, and the answer stays in their own prompt.

Will it ever approve something without me?

Only through a rule you wrote yourself in Settings ▸ Rules. A new rule starts out watching and answers nothing, the live command is checked again before anything is allowed, and every firing is written down with the rule's name.

Does it send my code or prompts anywhere?

No. There is no telemetry and everything stays on your Mac. The only network calls are the update check against GitHub Releases and, if you switch it on, Claude's usage quota.

What happens when AgentBar is not running?

Nothing breaks. The hooks never block the agent: if AgentBar does not answer, the agent falls back to its normal prompt in the terminal.

Which Macs does it run on?

macOS 12 or later, in the menu bar on any Mac, or as a Dynamic Island under the notch. There is also a CLI for Linux and a native tray app for Windows.

How do I remove it from an agent?

Flip that agent off in Settings ▸ Agents. AgentBar shows what it will take out and removes only its own entries; Show changes lists every edit it ever made, as a diff.

Install it in a minute.

One app and the hooks it installs, every change shown to you. No account, no telemetry.

View on GitHubDownload for macOS
$ brew install --cask michalstrnadel/tap/agentbar

Free & open source · macOS 12+ · No telemetry · v1.56.0