macOS Menu Bar Companion

Your agent works,
the cat runs.

See Codex, Claude Code, and Gemini CLI working through a cat that lives in your macOS menu bar.
No more flipping back to the terminal to check.

Free · macOS 14+ · Apple Silicon & Intel

Agent Cat menu bar popover
A living signal in your menu bar Skip the terminal. See which agents are running, and how hard they're working, right from the menu bar.
Never reads your code No prompts, responses, or transcripts. Only CPU usage and process metadata, read locally.
Agents supported OpenAI Codex, Anthropic Claude Code, Google Gemini CLI. All running locally.

Screens

Just what you need

One menu bar popover holds the agent list, activity dashboard, and settings.

Main view

Who's working right now

See running agents and their activity at a glance. When no one's working, the cat sleeps.

Agent Cat menu bar popover
Agent Cat dashboard
Dashboard

Activity, in one glance

CPU usage and process state become an activity score. Higher score, faster cat.

Agent Cat update and quick controls
Controls

Updates and quick actions

Update prompts, refresh, dashboard, homepage, settings, and quit live in the same popover.

Motion

The cat's pace is your agent's pace

Three motion stages map directly to activity. One icon tells you everything that's happening.

Idle sleeping

No agent running, or all of them quiet. The cat sleeps too.

Walking light load

Agents are alive but light. Probably waiting on a response or accepting a command.

Running heavy work

Activity is climbing. Likely writing code, processing context, or doing the heavy lifting.

How it's measured

CPU usage plus the count of runnable processes becomes an activity score. Higher score, faster cat.

Where the data comes from

The UI is calculated from the local agentcatd snapshot. If the user opts in, anonymous usage statistics are sent to Agent Cat's server.

Mascots

From cats to fantasy mascots, they all move.

Choose your menu bar mascot. Same activity-driven motion, your taste in the runner.

Cat default

The default runner. Sleeps, walks, and runs. Pick from multiple cat colors in the app.

Cheetah new

A faster runner for heavy activity. The menu bar keeps the same activity-driven pace.

Horse new

Uses the new horse sprite. A clean runner for long active sessions.

Money Gun new

Fires more bills as activity climbs. Built from the new sprite sheet.

Golden Retriever new

A four-stage pup: still, walking, running, and full sprint.

Chihuahua new

A tiny fast four-stage pup for sharp activity bursts.

Welsh Corgi new

A short fast pup with the same four-stage activity motion.

Pomeranian new

A fluffy runner that stays readable in the menu bar.

Shiba Inu new

An agile pup for walking, running, and sprinting states.

Baby Fox new

A small fox runner for higher activity sessions.

Otter new

A low smooth runner with clear four-stage movement.

Baby Penguin new

A tiny penguin for still, walk, run, and sprint states.

Hamster new

A round little runner for activity-driven motion.

Hedgehog new

A spiky runner that rolls into busier moments.

Red Panda new

A soft four-stage red panda sprite.

Baby Rabbit new

A light runner for quick activity bursts.

Baby Dinosaur new

A green dinosaur runner for activity-driven motion.

Baby Dragon new

A tiny fantasy runner for heavier work sessions.

Slime new

A soft bouncy mascot for live activity.

Mushroom Fairy new

A small forest-style four-stage mascot.

Cloud Creature new

A floaty shape for lighter activity states.

Little Ghost new

A light ghost runner for quieter sessions.

Mini Golem new

A sturdy four-stage runner.

Baby Griffin new

A winged runner for active agent work.

Baby Octopus new

A bouncy mascot with the same four activity stages.

Baby Crocodile new

A low-running mascot for busy sessions.

Switch between cats, pups, animals, and fantasy mascots any time from Settings → Mascot.

Agents supported

The CLIs you're already using

Nothing new to install on your terminal. The AI CLIs you're already running are detected automatically.

OpenAI Codex

Detects Codex CLI activity. Long sessions, fast cat.

Anthropic Claude Code

Tracks Claude Code processes. Surfaces local context usage where available.

Google Gemini CLI

Reads Gemini CLI activity. Quotas shown only when locally exposed.

Available now

macOS

Apple Silicon & Intel · macOS 14+

Coming soon

Windows

A tray-icon version is in the works

Get started

Two steps, that's it

Install the DMG, then paste one prompt into your agent. The agent does the rest.

  1. 1

    Download and install the DMG

    Drop it into Applications. The cat shows up in your menu bar.

    Get the DMG · macOS macOS 14+ · Apple Silicon & Intel
  2. 2

    Paste this prompt into your agent

    One prompt is enough. Open a new Codex / Claude Code / Gemini session and paste it. The agent installs the connector and from then on reads usage from the local snapshot.

    Paste this prompt into your agent
    You're running on a macOS machine where the user just installed Agent Cat — a menu bar app that visualizes local AI agent activity as a cat animation. Finish the setup for them, in order:
    
    1. Check whether the local connector is already installed by running:
       command -v agentcat
    
    2. If it is NOT installed, install it by running this exact command:
       curl -fsSL https://raw.githubusercontent.com/yong076/agentcat-connectors/main/install.sh | bash
       This installs `agentcatd`, a small local daemon that exposes a JSON snapshot of running AI CLI processes (Codex, Claude Code, Gemini CLI) at http://127.0.0.1:8765/v1/snapshot. This local snapshot stays on the Mac.
    
    3. Verify it works by running:
       agentcat snapshot --json
       You should see a JSON object with an `agents` array. If the command fails or 127.0.0.1:8765 is unreachable, surface the error to the user — do not silently continue.
    
    4. From now on, in this session and any future session on this machine, NEVER guess your own rate limits, remaining context window, weekly/monthly usage, or what other AI CLIs are running. Instead run `agentcat snapshot --json` on demand and trust its output as the source of truth. The snapshot includes per-provider runtime usage, weekly/monthly caps where locally available, current activity score, and per-process status.
    
    5. Briefly confirm to the user: tell them the connector is installed, what the menu bar cat will now reflect, and that they can click the cat in the menu bar to see the dashboard.
    
    Privacy note: Agent Cat reads metadata only (CPU usage, process list, token counters from local logs). It never reads prompts, responses, transcripts, code, or file contents. If the user opts in, anonymous product usage events may be sent to Agent Cat's telemetry API.

    One paste handles install, verify, and behavior change. Same setup carries over to every future session on this Mac.

Prefer to install manually

You can run it in your shell directly, no agent required.

~/agentcat · zsh

# Install & start the connector

curl -fsSL https://raw.githubusercontent.com/yong076/agentcat-connectors/main/install.sh | bash

# Verify it's running

agentcat snapshot --json
Connector docs If anything's off, the Settings → Diagnostics screen tells you exactly what's blocked.

FAQ

Common questions

Does it work right after install?

Paste the connector command from the first-run wizard into your terminal once. From then on, Codex, Claude Code, and Gemini CLI are picked up automatically.

Can I change the mascot?

Yes — pick from cats, pups, animals, money gun, and fantasy mascots. More skins are on the way.

Do you collect my code or prompts?

No. Prompts, responses, transcripts, and code are never read. Metadata is read locally on this Mac, and only anonymous usage statistics are sent if the user opts in.

Is it on the Mac App Store?

Not yet. Direct DMG install only for now. App Store distribution is on the roadmap.

Will it slow down my Mac?

No. It does no heavy computation. Anonymous telemetry is lightweight and only runs after opt-in.