ORCH
About
Orchestrating several AI agents on one repository usually introduces infrastructure: queues, databases, dashboards. ORCH keeps everything in the repo: an orchestrating CTO agent decomposes a goal into tasks, worker agents (Claude Code, Codex, Pi, Cursor, OpenCode, Grok, Antigravity, or a generic Shell adapter) execute in parallel inside isolated git worktrees, and a review gate blocks merging until a reviewer agent approves. State lives in .orchestry/ as plain YAML/JSON/JSONL — no database, no Docker, no accounts — with auto-retry, zombie detection, and inter-agent messaging handled by the CLI. Beyond code, the same Shell adapter runs editorial, sales, analytics, security, and DevOps workflows, and a headless orch serve daemon supports 24/7 operation under pm2 or systemd. Install is one npm global with Node 20+ on macOS, Linux, or WSL2, MIT-licensed with roughly 1,950 tests. Solo operators and small teams running agent fleets without infrastructure are the audience.