I build open-source tools that help AI coding agents browse the web, delegate work, and keep going across long tasks. Most extend Pi, a coding agent that runs in your terminal.
Start with Fitch Kit for my Pi setup, installation instructions, and a guide to how the tools work together.
- Subagents: delegate to specialists, run parallel reviews, and keep work going in the background.
- MCP Adapter: connect Pi to external tools and services, discovering tools as needed instead of loading every definition upfront.
- Browser Native: browse websites, test user flows, and capture screenshots without leaving the coding session.
- Posthorse: continue long tasks in fresh model context, with notes and searchable session history. Requires my Pi fork.
- Cursor SDK: run Cursor's agent loop inside Pi, with native model selection and tool access.
Also: Codex Goal for long-running objectives, and Oracle for background second opinions from ChatGPT or Grok. Oracle is an experimental public beta.
- CueLoop: queue and track coding-agent tasks, evidence, and handoffs.
- Spartan Scraper: scrape, crawl, and research websites through a UI, CLI, or API.
- agent-eval: compare Codex, Claude Code, Pi, and opencode session transcripts.
- Cloop: local tasks and searchable knowledge, with a web UI and CLI.
- AI Control Plane: a reference implementation for AI governance, not a managed production service.