English | 简体中文
A local-first AI desktop companion with transparent Markdown memory.
It reacts to your keyboard, mouse, and controller, supports chat, and keeps persistent companion memory in editable local files.
Quick Start · Download · Highlights · Local Memory · Contributing
Based on ayangweb/BongoCat, this fork pushes the project toward a more companion-like direction: not just a cute desktop pet, but a pet that can chat, remember, and stay with you over time.
Most AI products still live inside a chat box.
BongoCat AI takes a different direction: a local-first AI desktop companion with a persistent memory layer that stays close to the desktop instead of hiding behind a web app.
It keeps the original BongoCat feeling — lightweight, cute, responsive — and extends it with:
- real-time keyboard, mouse, and gamepad reactions
- a dedicated companion chat window
- editable Markdown memory with persistent context
- desktop-native controls like tray, startup, always-on-top, and window behavior settings
- Live2D desktop pet rendering with transparent background
- Real-time reactions to keyboard, mouse, and controller input
- Window size, opacity, radius, position, and click-through controls
- Tray controls and desktop-friendly behavior for long-running use
- Dedicated companion chat window
- Configurable provider, model, base URL, and API key
- Streaming-style replies and multi-turn context
- Designed around companion UX instead of pure tool-style interaction
- Memory lives in editable local Markdown files
PERSONA.mdfor character and toneMEMORY.mdfor durable memory and stable preferencesdaily/YYYY-MM-DD.mdfor recent context and short-term notes- Inspired by the same transparent, local-first memory direction that made projects like OpenClaw resonate, without hiding state behind a closed memory layer
- Tauri 2 + Vue 3 + TypeScript + Rust + Live2D
- Native tray, startup, shortcut, and window integration
- Windows, macOS, and Linux (X11) support
- Distributed through GitHub Releases
| macOS | Windows | Linux |
|---|---|---|
 |
 |
 |
- Download the latest release:
- If you need package names for your platform:
- Launch the app and try:
- the desktop pet window
- input reactions
- tray controls
- chat window and AI settings
If you want a more personal setup, configure your model, AI provider, and memory settings in Preferences.
The original project already delivers a polished desktop pet with Live2D rendering and input reactions.
This fork adds a second layer on top of that foundation:
- companion chat, not just animation
- persistent memory context, not just one-off replies
- editable local memory files, not invisible black-box storage
The goal is simple: turn a desktop pet into a local-first AI companion layer rather than just a themed chat UI.
BongoCat AI uses a local-first Markdown memory workspace to keep companion context persistent, transparent, and user-editable.
Current memory files are stored like this:
<appData>/memory/
├── PERSONA.md
├── MEMORY.md
└── daily/
└── YYYY-MM-DD.md
PERSONA.md: personality, tone, and character directionMEMORY.md: long-term facts and stable preferencesdaily/YYYY-MM-DD.md: day-specific context and temporary notes
- Markdown is the source of truth
- Core memory stays local and visible
- Users can inspect and edit what is remembered
pnpm installpnpm tauri devpnpm tauri build- Live2D desktop pet with multi-input reactions
- Chat window and base AI settings
- Local Markdown memory MVP
- Transparent memory file access from settings
- More robust structured memory extraction
- Better retrieval from historical daily notes
- Emotion and proactive behavior systems
- Richer environment awareness
- developers who want a desktop pet with real interaction
- users who like cute UI but still want practical AI features
- people interested in local-first memory instead of opaque AI state
- builders exploring the overlap between desktop pets, agents, and companion UX
If you like the idea of combining desktop pets, AI, and local memory, contributions are welcome.
- Read the Contributing Guide
- Open an issue or pull request
- Share ideas for models, interactions, and companion behavior
This project is based on the upstream MIT-licensed project and remains compatible with that license. See LICENSE.