Voice-first AI Agent for Android — not a chat wrapper.
English | 中文
Rocky organizes voice interaction, task execution, system bridging, and result review into a native Android agent experience.
- Voice-first — voice is the primary interface, not a chat list
- 30+ native Android tools — contacts, calendar, weather, location, reminders, alarms, camera, browser, crypto, and more
- Multi-provider AI — OpenAI, Anthropic, Gemini, Azure, Groq, xAI, OpenRouter, DeepSeek, Doubao, aiProxy
- Realtime voice — live sessions via OpenAI, Gemini, and Doubao realtime APIs
- Custom skills — built-in skills plus user-importable custom skills
- Local execution — on-device shell and Python 3.11 runtime via Chaquopy
- Characters & Souls — configurable AI personality and voice
- Memory — persistent context across sessions
Voice Input → Voice Engine → AI Provider → ROS Runtime → Execution Layer → Results → UI + Voice
Rocky is built around the ROS (Rocky OS) Runtime, which manages sessions, tools, skills, voice bridges, characters, and memory. Execution happens across three layers:
| Layer | Description |
|---|---|
| Android Native Bridge | Kotlin code calling system APIs (contacts, calendar, location, etc.) |
| AI Tool Layer | Actions dispatched through provider APIs |
| Local Execution | Sandboxed shell/Python runtime via Chaquopy |
Providers follow a three-layer abstraction: Provider → Account → Model.
git clone https://github.com/openrocky/openrocky_android.git
cd openrocky_android
./gradlew assembleStandardDebugFor full build instructions, project structure, and code style, see the Development Guide.
Contributions are welcome! Please feel free to submit a Pull Request.
- Website — project homepage
- Discord — chat with the community
- Telegram — @openrocky
- iOS version — OpenRocky for iOS
Built by @everettjf with Claude Code and Codex.