A 5th Claude Code instance watches, coordinates, and dispatches work to 4 parallel agents. It learns what works, evolves its playbooks, and gets better every session.
$ npx ninja-terminals
A master orchestrator watches your 4 agents, dispatches tasks, detects when they're stuck, and learns from every session. You describe the goal — it figures out the rest.
A 5th Claude Code instance coordinates the other four. It reads their output, detects status changes, dispatches tasks, and intervenes when things go wrong — via browser automation.
Research, build, test, and verify — all running simultaneously in a browser-based terminal grid. Each agent has its own scope, its own context, and real-time status detection.
The orchestrator measures tool effectiveness, evolves playbooks, and rates what works — based on real session data. It gets smarter every time you use it.
No configuration files. No orchestration boilerplate. Just describe what you want to build.
One command spins up a local Node server and opens Ninja Terminals in your browser. Your existing Claude API key and MCP configs are auto-detected.
npx ninja-terminals
A 5th Claude Code instance — the orchestrator — reads your goal, breaks it into parallel subtasks, and dispatches each one to a terminal via browser automation.
Four agents run in parallel while the orchestrator monitors them — detecting status changes, unblocking stuck agents, and re-dispatching work. You watch. It coordinates.
The core is free and open source. Pro is a one-time purchase — no subscriptions, no seats, no vendor lock-in.