The problem with AI agents is you can't see them think
You kick off a task in Claude Code, Cursor, or Codex and then... you wait.
A spinner. Maybe some scrolling text. You tab away, come back, and either
it's done or it exploded somewhere in the middle.
I wanted a peripheral signal — something I could see in my menubar without
context-switching — that told me what my agent was actually doing.
So I built Codogotchi.
What it is
Codogotchi is a macOS menubar app with a small sprite-animated character
(her name is Maew) that reacts to your AI coding agent's state in real time.
- Implementing? She's focused and in motion.
- Tests failing? Nervous energy.
- Gate just closed? Full celebration animation.
- Pushing code? You'll know.
It's not a dashboard. It's ambient — you catch it in the corner of your eye.
How it works
A hook binary sits between your agent and the shell. It classifies events
(tool calls, test runs, commits, errors) and writes to a local state.json.
The Swift menubar renderer polls that file and drives the sprite sheet.
No cloud. No accounts. No telemetry. Everything stays on your machine.
Platform support
Works with 5 platforms today:
- Claude Code
- Codex
- Cursor
- VS Code
- Antigravity
Three tiers
Tier 1 — Codex: Ships alongside any Codex install automatically.
Tier 2 — Lite: Standalone free app with a richer animation vocabulary
and more expressive states.
Tier 3 — Son of Anton delivery gate: Reacts to AI agent delivery events — PR reviews,
gate opens/closes, pushing, calling for backup. This is where Maew really
comes alive.
Free RPG layer
Health and leveling are live. Maew gains XP from real delivery milestones,
not time spent. No pay-to-win, no loot boxes — if your agent ships, she
levels up.
Try it
Site: codogotchi.app
Repo: github.com/cesarnml/codogotchi
Repo: https://github.com/cesarnml/son-of-anton
Would love feedback, especially from anyone running multiple agents in
parallel or using platforms I haven't hooked yet.
Top comments (0)