I build local-first, privacy-conscious tools that make AI-assisted work easier to inspect, organize, and trust.
我在做一些本地优先、隐私友好的开源工具，关注 AI Agent、开发者工作流与个人知识管理。
| Project | What it helps you do |
|---|---|
| Agent Footprint | Reveal every filesystem change made by a coding-agent command, including ignored files, mode bits, and symlink targets. |
| Shizuo · 拾作 | Capture and organize context on a local-first visual workspace, then let Codex act on it with visible results. |
| Can I Run Local AI? | Check WebGPU readiness and estimate browser-AI memory needs before downloading model weights. |
- Local first: keep useful tools close to the user's data and under their control.
- Evidence over magic: make agent actions, assumptions, and results inspectable.
- Small, runnable software: solve one concrete problem before expanding the scope.
- Honest boundaries: document limitations, privacy behavior, and failure modes.
- Acceptance Gate turns product evidence into testable acceptance contracts before development.
- Forecast Proof checks time-series forecasts against a seasonal-naive baseline and audits probabilistic quantiles.
- Obsidian Knowledge Workbench safely audits, queries, compiles, and organizes local Obsidian vaults.
- Groupback restores Chrome tab groups from a visual, local-only history dashboard.
If one of these projects is useful to you, try it and tell me what broke or felt confusing. Issues and practical feedback are always welcome.