GitHub radar
Watch your AI coding agents and approve permissions in one click
Coucou is a little mascot on your screen that watches your AI coding agents and lets you approve their actions without switching to the terminal.
It's a small mascot app that shows what your AI coding agents are doing and lets you approve or deny their requests right on screen.
When it helps
Useful if you run several agents (Claude Code, Cursor, Codex and similar) and are tired of switching to the terminal to confirm actions; skip it if you use one agent and confirmations aren't a hassle.
Pros
- Approve or deny agent requests with a single click
- See in real time which files and commands are changing
- Approve permissions from your phone with Face ID
- Open source under MIT — you can check how it handles your data
Cons
- Full feature set (offline dictation, call notes) is macOS-only
- Voice input and call transcription need your own Anthropic API key
How to set it up — step by step
- 1Install Coucou: on Mac run brew install --cask coucou; on Windows or Linux download the installer from the repo's releases.
- 2Launch the app and open your AI coding agent (for example, Claude Code) in your project folder.
- 3Send your agent the text from the block below so it shows you what a permission request looks like in the widget.
- 4Check that you can see the list of files and commands your agent is changing right on screen.
- 5If you want voice input or call transcription, get your own Anthropic API key ready for that feature.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install Coucou from https://github.com/Louis-CFM/coucou — on Mac via brew install --cask coucou, on Windows or Linux using the installer from the releases page. Start it and show me what a Claude Code permission request looks like in the widget. If you want to turn on the voice assistant or Claude-based call transcripts, you'll need my Anthropic API key — ask me for it.
Light: the app itself is small; a Mac, Windows or Linux machine all work. The full set — offline dictation and call notes — is Mac-only for now and needs Apple Silicon (the speech model is about 650 MB); Windows and Linux already get agent-watching and most other features.
Open on GitHub▌ More finds