Organized AI / Ecosystem Research
A production TypeScript monorepo for an entire AI agent ecosystem. Seven packages: unified LLM API, agent runtime, coding agent CLI, Slack bot, terminal UI, web UI components, and GPU pod manager. Each package is independently publishable but shares a root tsconfig, biome linter, and npm workspace setup.
Four purpose-built CLIs: Steer (Swift GUI automation via OCR), Drive (Python tmux control with sentinel pattern), Listen (FastAPI HTTP job server on port 7600 spawning Claude Code), and Direct (CLI client for Listen).
A self-improving Python agent with a closed learning loop — creates skills from experience, improves them during use, maintains persistent memory, runs FTS5 session search for cross-session recall. Ships with cron scheduler, multi-platform messaging gateway (Telegram, Discord, Slack, Signal), and explicit OpenClaw migration: hermes claw migrate.
Node.js server + React UI orchestrating AI agents to run a business. Org charts, heartbeat scheduling, per-agent budgets, multi-company isolation, full tool-call audit logs. Uses pnpm workspaces across packages/, server/, ui/, cli/, skills/. Works with OpenClaw, Claude Code, Codex, Cursor, Bash, HTTP.
Cross-repo synthesis