DevLab is the public entry point for AI work I build, maintain, or actively contribute to around four themes: AI coding workflow, token analytics, agent security, and ChatGPT workflow performance.
I use this repository as a documentation-first front page for current AI projects and as a durable home for earlier multi-language experiments. The goal is practical: show what I am building, what I am maintaining, and how I approach AI developer tooling in real repositories.
| Project | Role | Domain | Stack | Status |
|---|---|---|---|---|
| Token Insight / Repo | Builder / Maintainer | Local-first token observability for AI coding tools | Rust, React, SQLite | Active build |
| Codex Composer / Repo | Builder / Maintainer | Reproducible Codex workflow bootstrap for repositories | JavaScript, Shell, Markdown | Active iteration |
| ChatGPT TurboRender / Repo | Builder / Maintainer | Keep long ChatGPT conversations responsive without replacing the native UI | TypeScript, WXT, Manifest V3, Playwright, Vitest | Active build |
| AgentScan / Repo | Contributor / Maintainer | Exposed AI agent discovery and security audit | Go, React, SQLite | Active security tooling |
- Role: Builder / Maintainer
- Domain: Browser-side ChatGPT long-thread responsiveness
- Stack: TypeScript, WXT, Manifest V3, Playwright, Vitest
- Status: Active build
This project keeps long ChatGPT sessions responsive by reducing browser render pressure instead of replacing the native UI. It trims cold history before first render, preserves a hot interaction window, and restores old turns on demand so long conversations stay usable.
View details:
- one-api: maintenance-oriented fork work around LLM gateway and API distribution scenarios.
- MetaGPT: participation via fork-based exploration of multi-agent software workflows.
- Cookbook Index
- How I Think About Codex Workflow Automation
- Codex Composer development notes
- Token Insight local-first analytics notes
- AgentScan engineering notes
Earlier experiments remain available as reference code:
I am interested in collaborations around AI developer tools, internal AI workflow enablement, local-first analytics, and agent security. DevLab is structured to make that evaluation quick: projects, notes, and older experiments are all linked from one place.