I built a free local app that runs a team of agents

Description and use

RandComp Studio is a local app that coordinates multiple AI agents, visualizing tasks and outputs without requiring an account or external server.

Problem

Managing multiple Claude Code sessions simultaneously caused confusion about task ownership and progress.

Outcome

RandComp Studio enables local, account‑free coordination of multiple AI agents, visualizing tasks and outputs while allowing interaction through app, CLI, or MCP.

Workflow

• Describe a project to RandComp Studio
• The coordinator agent splits the project into tasks
• For each task, a helper agent (developer, QA, designer, etc.) is started as a separate Claude Code session in its own terminal
• The app visualizes the growing task diagram, outputs, and pending user actions
• The user interacts via the app, CLI, or MCP to monitor progress
• Optional: use Codex, Kimi, or local models via Ollama

Evidence

I kept running several Claude Code sessions at once and losing track of who was doing what and what was waiting on me.

Source

So I built RandComp Studio: you describe a project, a coordinator agent splits it into tasks and starts helpers (developer, QA, designer…), each a real Claude Code session in its own terminal.

Source

The app shows a diagram of the tasks as it grows, what each one produced, and what needs you.

Source

You can use it three ways: the app, a randcomp CLI, or MCP so Claude Code itself can ask how the team is doing.

Source

It also works with Codex, Kimi and local models via Ollama.

Source

Local only: no account, no server of mine — it runs the CLIs you already pay for.

Source

It sends four anonymous usage counts you can switch off in Settings.

Source

Free early preview, Apple-silicon Macs for now (Windows/Linux later — one-person army).

Source