AI agent developer. I build local-first agents that can break down tasks, use tools, stream their work, retain useful context, and turn repeated work into automation β on infrastructure you can run and inspect yourself.
Years of production engineering across JVM services and Rust systems shape this work. I now build the ecosystem below in tight implementation and review loops with the agents themselves: Bodhi helps build Bodhi.
These are the main projects I am actively developing:
| Project | What it is |
|---|---|
| π§ Jiandu | Small, agent-independent, filesystem-backed memory: a Rust library plus one stdio MCP memory tool for Session, Project, and Global memory. |
| π Bamboo-agent | A local-first AI agent runtime in Rust. It owns sessions, tools, skills, sub-agents, workflows, schedules, and context policy behind HTTP and streaming APIs. |
| π§ Bodhi-AI | The desktop agent app: a Tauri shell around the Bamboo runtime and Lotus interface. |
| πͺ· Lotus | The React interface for live reasoning, tool calls, tasks, projects, and agent control. lotus-next is its mobile-first companion track. |
| β¨ Nova | A self-contained Rust MCP server for macOS screenshots, Set-of-Mark targeting, OCR, mouse, and keyboard control. |
| π Zenith | The workspace that coordinates the repositories, architecture, roadmap, and release train. |
| π― Pavilion | The bilingual δΈζ/English website and long-form product and architecture documentation. |
- Local-first and private β your model keys, your data, and your machine; cloud services are optional.
- Transparency over magic β make model output, tool calls, state changes, and failures visible.
- Shared memory with a narrow boundary β durable filesystem memory through a small MCP surface, with product-specific context policy kept in the agent runtime.
- Real work, not just chat β task decomposition, sub-agents, schedules, workflows, and desktop control.
- Agent-native development β parallel implementation, adversarial review, exact acceptance checks, and automated follow-through as daily practice.
- Clear implementation boundaries β Rust for local execution and durable services; MCP where different agents need a shared interface.
Daily drivers: Rust Β· TypeScript/React Β· Tauri Β· Go Β· MCP Β· HTTP/SSE/WebSocket
Also fluent in: Java Β· Kotlin Β· Scala Β· Vue.js
Feel free to reach out at mugeng.du@gmail.com.