A desktop pet that reacts while your AI coding agents work , and grows the more you code.
Website · Install · Pet gallery · Leaderboard · Discord
If AgentPet brightens your workflow, please give it a star , it genuinely helps others find it.
English · Tiếng Việt · 简体中文 · 日本語
Run several coding agents at once (Claude Code, Codex, Gemini, Cursor and more) and it gets hard to tell who is working, who is done, and who is waiting for your input. AgentPet answers that at a glance: a menu-bar monitor shows every agent's live state, and a little pixel pet floats on your desktop and reacts to it all.
It is also a tamagotchi. The pet is fed by real work , the tokens your agents burn and the sessions they finish , so it earns XP, levels up, unlocks achievements, and evolves as you code. Free, open source, and fully offline unless you choose to connect.
It is built and given away for free. A coffee (or a star) keeps the features coming.
- Highlights
- Features
- Supported agents
- Screenshots
- Install
- Getting started
- Usage dashboard
- Pets
- Platforms
- Community ports
- Contributors
- Acknowledgements
- Support
- License
- 🐾 Ambient desktop pet that mirrors your agents' state without breaking your focus.
- 📊 Menu-bar monitor with every agent, its project, what it's doing, and a live timer.
- 🎮 Raise your pet: real tokens and finished sessions feed it, level it up, and unlock 14 achievements.
- 🗂️ Per-project pets and a per-project token usage dashboard (daily and monthly).
- 🔌 11 coding agents supported out of the box, plus a universal wrapper for anything else.
- ☁️ Optional web profile, leaderboard, and cross-device restore , everything works offline if you skip it.
- 🌍 Localized in English, Tiếng Việt, 简体中文 and 繁體中文.
- 🍎 macOS (native Swift/SwiftUI, notarized) and 🪟 Windows (Tauri), both with auto-update.
Monitor your agents
- Multi-agent monitor in the menu bar: a live list of every agent with a colored status dot, the project, what it's doing (running tool / reason it's waiting), and a per-state timer counting in real time.
- At-a-glance icon: shows how many agents are running, and turns orange with a count when one needs your input.
- Native notifications when an agent finishes or asks for input, with sounds you can toggle.
- Reactive, themed speech bubbles per agent, with a model badge, plus fully custom messages and appearance if you want them.
- Session history: the last 90 days of sessions are kept locally so you can look back.
Raise your companion (tamagotchi)
- Your pet eats the real tokens your agents burn and the sessions they finish (including subagents). It earns XP, levels up, and evolves through five stages: Hatchling → Companion → Scout → Hero → Legend.
- Every pet keeps its own level, so the more you code with a companion, the more it grows.
- 14 achievements from real milestones: first session, level-ups, token burns (1M / 10M / 50M), session counts, feeding streaks, a Night Owl badge, and more.
- Stats HUD: right-click the pet for a game-style card , level, XP, hunger, a 7-day burn chart, your streak, and your live Claude / Codex subscription limits read directly from your sign-ins (no extra tool needed).
- Break reminder (off by default) nudges you to step away after long stretches.
Work across projects
- Per-project pets: assign a companion to a project folder, and that project's XP goes to it.
- Split pet: configured projects get their own on-screen pet; everything else shares your main one.
- Usage dashboard: see exactly where your tokens go, per project and per agent, day by day and month by month (see below).
Share and sync (all optional)
- Web profile & leaderboard: sign in with GitHub to show your companions at agentpet.thenightwatcher.online and climb the community leaderboard by level, sessions, or tokens.
- Cross-device restore: sign in on a second machine (or after a fresh install) and your pets come back at the level you raised them. Progress merges forward only, so it is never undone on either machine.
- Community pet gallery: browse thousands of pets, adopt one in a click, submit your own, and request new ones.
AgentPet runs completely offline. Connecting only adds the web profile, leaderboard and cross-device sync , nothing leaves your machine until you sign in.
Install a hook from Settings with one click, or wrap any command.
| Agent | State detail |
|---|---|
| Claude Code, Codex, Gemini CLI, opencode, Factory Droid | working · waiting for input · done |
| Cursor, Windsurf, Antigravity, GitHub Copilot, Kiro CLI, Pi | working · done |
| GLM (Z.AI) | works through Claude Code automatically |
| Anything else | agentpet run -- <command> , working while it runs, done when it exits |
Right-click the pet for a game-style HUD , level, XP, hunger, a 7-day burn chart and your live Claude/Codex limits.
![]() Menu-bar monitor , every agent at a glance |
![]() Care tab , raise your pet with real work |
![]() Native, tabbed Settings |
![]() The desktop pet |
brew install --cask ntd4996/tap/agentpetGrab the latest AgentPet.dmg from Releases, open it, and drag AgentPet to Applications. Builds are Developer ID-signed and notarized by Apple, so they open without a Gatekeeper warning, and update themselves via the menu-bar Updates button.
Download the installer or portable build from the website or the releases. The first launch may show a SmartScreen warning (the Windows build isn't code-signed yet); click More info → Run anyway. It installs per-user (no admin) and auto-updates.
git clone https://github.com/ntd4996/agentpet.git
cd agentpet
./scripts/build-app.sh release
open build/AgentPet.appRequires Xcode 16 / Swift 6. The Windows app lives under windows/ (Tauri + Rust).
- Launch AgentPet , it lives in the menu bar.
- Open Settings → General and click Install next to Claude Code (or your agent), then Enable notifications.
- Run an agent. The pet reacts, and the menu-bar icon shows who is working or waiting.
- (Optional) Open Settings → Care → Sign in with GitHub to sync your companions and unlock the web profile, leaderboard and cross-device restore.
- In Settings → General, click Remove next to each connected agent (this strips AgentPet's hooks so the agents don't error after the app is gone).
- Remove the app and its data:
brew uninstall --cask agentpet # or drag /Applications/AgentPet.app to Trash
rm -rf ~/.agentpet # downloaded pets + local state
rm -f ~/Library/Preferences/com.agentpet.app.plistSign in and open the Usage dashboard (in your profile menu on the website) to see where your tokens actually go:
- By project and by agent, so you can tell which repo and which model is burning the most.
- Daily and monthly views, with a chart and per-project / per-agent breakdowns.
- Only the last folder name of each project is ever sent, never the full path, and nothing is logged until you connect.
Admins get a community-wide view plus a per-user drill-down.
Pets use the open Codex pet-pack format (pet.json + a spritesheet). You can:
- Browse the online library and adopt a pet in one click (Settings → Pet → Browse pets, or the web gallery).
- Map animations to states, resize, rename, and customise chat lines.
- Make your own and submit it to the community gallery.
A starter pet is installed on first launch. AgentPet bundles no pet art , every pack is added at runtime, and each asset is owned by its submitter under their own license.
- macOS 13 Ventura or later , native Swift / SwiftUI, notarized, Sparkle auto-update, Homebrew cask. Apple Silicon and Intel.
- Windows 10 / 11 (64-bit) , Tauri (Rust), auto-update. Feature parity with macOS.
Under the hood: a Unix-socket daemon receives agent events from lightweight hooks, an on-device store drives the pet, and an optional Astro + Cloudflare Workers/D1 backend powers the web profile, leaderboard and usage dashboard. See docs/specs for the design.
AgentPet is built for macOS and Windows, but the community has reimagined it elsewhere:
- Linux (Rust + GTK4) , agentpet-linux by @tranhuuhuy297. An independent, from-scratch port for Ubuntu (Claude Code + Codex).
These are separate community projects, not maintained here. Building one? Open an issue and we'll link it.
AgentPet is a community effort. Huge thanks to everyone who has shipped code, pets, translations and ports , see the full Hall of Fame.
- @ntd4996 , creator and maintainer
- @hoangphison , co-founder
- ...and many more contributors on the website and in the repo's contributors list.
Contributions are welcome , features, agent integrations, pets, translations and bug fixes. Open an issue or a pull request.
AgentPet stands on the shoulders of an open community.
- The idea and the pet-pack format come from Petdex (MIT). AgentPet is an independent, interoperable client: it reads packs in Petdex's format and can adopt them from Petdex's public API.
- The pets themselves are the community's. The gallery mirrors thousands of companions from Petdex and OpenPets, alongside pets submitted directly by AgentPet users. Every pet asset is owned by its respective creator under their own license.
- Contributors and community ports (above) shaped much of what AgentPet is today.
AgentPet bundles no pet art of its own. If you hold rights to a character and want it removed, please see our takedown page, and for pets sourced from Petdex, direct takedowns there.
If AgentPet saves you some tab-hunting:
- ⭐ Star the repo so more people find it.
- ☕ Buy me a coffee or support on Ko-fi to fuel more features.
Built by Nguyễn Thành Đạt (@ntd4996) and the community.
MIT , see LICENSE. Application code only; pet assets are not part of this repository and belong to their respective creators.





