I build software that connects LLMs with real systems — local tools, desktop automation, native macOS utilities, and polished end-user apps.
| Project | What it is | Stack |
|---|---|---|
| Pipi-Shrimp Agent | Local-first AI desktop agent with multi-provider LLMs, tool calling, browser automation, workflows, project memory, multi-agent collaboration, and Typst rendering. | Tauri · Rust · React · TypeScript |
| Conveyor | Self-hosted control plane for running Codex from Telegram or Feishu, with isolated git worktrees, job queues, auditability, and optional Mac execution nodes. | Python · SQLite · Codex · Telegram/Feishu |
| Cool Down Pro | Native macOS thermal and fan-control utility using SMC, a privileged helper, sensor fusion, hysteresis, asymmetric EWMA filtering, and cooldown logic. | Swift · XPC · SMJobBless · IOKit |
| Capture Your Screen | Native macOS screenshot + annotation app with a deliberately read-only ScreenCaptureKit CLI for automation agents. | Swift · SwiftUI/AppKit · ScreenCaptureKit |
| 阅笺 / Yuejian | Local-first, ad-free Android TXT/EPUB reader focused on large-file performance, TTS, offline reading, and a polished Compose experience. | Kotlin · Jetpack Compose · Material 3 |
| Resume Generator | Full-stack resume workspace for importing, editing, AI-assisted tailoring, and Typst-backed PDF/SVG export. | React · TypeScript · Node.js · Typst |
- Local AI agents that can safely use real tools instead of living only in a chat box.
- Native system software where OS permissions, IPC, sensors, performance, and reliability matter.
- Product-quality apps with clear UX, packaging, release workflows, tests, and documentation.
- Small systems with explicit trust boundaries — read-only observation, privileged helpers, isolated worktrees, confirmations, and kill switches where appropriate.
LLM / Agent layer
↓
Tools + orchestration + memory
↓
Native / OS integration
↓
Real user workflows
I care about the parts between the demo and the product: failure recovery, security boundaries, packaging, persistence, performance, and making complex systems understandable to users.
I’m currently pushing deeper into the intersection of AI agents + native desktop systems: local execution, computer-use safety, developer automation, and practical open-source tools people can actually install and use.
Open-source projects, native apps, and experiments live here. Most flagship repositories include architecture notes, screenshots, release instructions, and implementation details.