Gabo

A multi-agent harness that runs on your own machine.
A team of twelve role agents works together in rooms built for studying, brainstorming and shipping software.

---

Gabo is a local web app built on the [Claude Agent SDK](https://code.claude.com/docs/en/agent-sdk/overview). It works like a terminal coding agent (tools, permissions, slash commands, history, plugins, skills, MCP servers), and adds a team of role agents who hand work to each other, challenge each other and build on each other's output.

It runs on the AI account you connect: a **Claude**, **OpenAI** or **Gemini** API key, or a free **local model** in Ollama.

Features

• **Twelve agents with distinct roles:** Believer, Skeptic, Investor, Judge, Designer, Coder, Tester, Researcher, Tutor, Caveman, Planner and Emperor. Each has a pixel mascot and its own stage.

• **Rooms where the team works together:**

  • Library (study and research)
  • Arena (brainstorm and pick the best idea, with an idea tournament)
  • Hackathon (plan, design, build, test, ship)
  • Laboratory (you pick the team)

  In every room, agents answer each other's points and pass work along, and the chat shows who builds on whom.

• **Team size that fits the prompt:** a router picks a tier with no model call. It scores the prompt: things you ask it to produce (a plan, questions), amounts ("3 examples"), several steps ("then…"), lookups and length. Plain short questions are Quick.

  • Quick: short questions. The team answers together in one reply.
  • Standard: regular tasks. Like Quick, with at most one real agent when a role needs tools.
  • Deep: build work, long tasks, the Arena tournament. Real agents, with independent ones running in parallel.

  A team-in-one-reply answer used about half the tokens of a real-agent run