Stack
Case Study
Beta
orca — CLI Agent Orchestrator
A single-file Python CLI I built to fan work out to parallel headless AI worker agents — Claude Code plans and delegates, cheap models do the writing. Open source, pipx-installable.
Context
Tags
Overview
orca is a single-file, dependency-free Python CLI that dispatches parallel headless AI worker agents and collects their results as durable on-disk artifacts (prompt, log, clean output, session id, status). I run Claude Code (Opus) as the orchestrator: the expensive model plans and hands out small instructions, and cheap open-source models running in workers do the actual writing.
It's how I do large jobs on a tiny budget — for example, modularizing ~14k lines of code that a single Claude session can't hold. Full write-up on the blog: My AI Coding Setup.
Features
Design
- Pluggable adapters for many worker CLIs (cmd, Cursor, Claude, Codex, Kimi, OpenCode) — contributors can add their own in a few lines.
- Event-driven via background shells: no polling, resume/fork support, telemetry per task.
- pipx-installable as an `orca` command; no runtime dependencies.