Superset – Orchestrate 100+ Coding Agents in Parallel

TL;DR

TL;DR: Superset is a desktop terminal app that orchestrates multiple AI coding agents (Claude Code, Codex, Cursor, etc.) in parallel worktrees, letting you manage 100+ agents simultaneously and automate code review workflows from one interface.

Source and Accuracy Notes

• GitHub: superset-sh/superset — 11.4k stars, Apache-2.0 license
• Official Site: superset.sh
• Tech Stack: TypeScript, Electron, TUI

What Is Superset?

Superset is a code editor and orchestration platform built for the AI agents era. It runs as a desktop app with a terminal-style interface where you can spawn multiple AI coding agents simultaneously, each in its own isolated worktree. Think of it as a mission control dashboard for an army of Claude Code, Codex, Cursor, and other agents — all accessible from one window.

The core value proposition is parallelism: instead of running one agent task at a time, Superset lets you fire up dozens of agents that each handle a different branch, PR review, file refactor, or automation task. The agents share the same codebase context but run in isolated git worktrees, so they do not step on each other’s changes.

It supports agents from:

• Claude Code (Anthropic)
• Codex (OpenAI)
• Cursor
• OpenCode
• GitHub Copilot
• Amp (Google)
• Gemini (Google)

Setup Workflow

Step 1: Download for macOS

Navigate to the official site and download the macOS binary:

open https://superset.sh
# Click "Download for macOS"

The app ships as a .dmg installer. After mounting and dragging to Applications, launch Superset from your dock or Spotlight.

Step 2: Configure Agent Paths

Superset auto-detects agents that are installed globally. If an agent is not on your PATH, you can configure custom binary locations per workspace.

# Verify your agent is accessible
which claude  # should return a path
which codex   # or wherever Codex is installed

In Superset, open Settings → Agents to verify detection. Each agent shows its version, status, and configured path.

Step 3: Create a New Workspace

Workspaces are isolated working directories, each with their own git worktree. To create one:

1. Click New Workspace in the sidebar
2. Choose a root directory like ~/projects/
3. Select which agents to enable in this workspace
4. Superset initializes a git worktree for each agent under ~/.superset/worktrees/

# Workspaces live at ~/.superset/worktrees/{workspace-name}/
ls ~/.superset/worktrees/

Step 4: Open New Tabs

Click New Tab or use the sidebar to spawn additional agent sessions — each tab runs its own agent instance in the workspace context.

Orchestration Features

Automations

Rather than manually triggering agents, you can define automations — scripted sequences that spin up agents, run tasks, and collect results. Example automation flows:

• Parallel branches: Create multiple worktrees from a feature branch, run agents on each to tackle different sub-tasks
• PR review: Route a PR through multiple agents simultaneously to get reviews from Claude, Codex, and Copilot
• Code diff analysis: Forward changed files to a set of agents for parallel review

Automations are defined in cloud-workspace.ts config files per workspace.

Port Forwarding for Live Preview

When agents run dev servers, Superset can automatically forward local ports:

# Superset tracks port usage per workspace
PORTS
5
use any agents        3002
see changes          3000
open in any IDE       3001
5678

Agents can forward ports so you can preview running services without manually managing webpack-dev-server, Vite, or Django runserver instances.

MCP Server Integration

Superset ships with an MCP server that connects to the Superset desktop app itself, enabling Claude Code and other agents to interact with the Superset workspace API directly:

MCP SERVERS
1 connected
config:.mcp.json

Agents use the .mcp.json config to connect to this MCP server for programmatic workspace control.

Git Worktree Branching

Each agent runs in its own git worktree. This is the key architectural decision:

# An agent working on feature/auth gets its own branch worktree
cd ~/.superset/worktrees/superset/feature-auth
git branch
# * feature/auth  (detached at abc123)

This means:

• No agent accidentally commits to main
• Changes are isolated until you merge
• Multiple agents can work on conflicting branches without data loss

Practical Evaluation Checklist

• [x] Workspaces with isolated git worktrees
• [x] Parallel agent execution (100+ simultaneously)
• [x] Automations for scripted multi-agent workflows
• [x] MCP server integration (.mcp.json)
• [x] Port forwarding per agent
• [x] Built-in code review panel (PR # Approve/Comment)
• [x] Dark terminal UI (Electron + terminal renderer)
• [x] Agent auto-detection on macOS
• [x] File browser and change diff viewer
• [x] Cloud + Mobile + CLI port routing

Security Notes

• Agents run locally under your user account — no cloud dependency for core functionality
• Git worktrees are created under ~/.superset/worktrees/ — ensure this directory has appropriate Unix permissions
• MCP server runs on localhost — only grant access to trusted local processes
• Superset does not send code to third-party servers; all agent communication stays between the agent SDK and your configured API keys
• Review .mcp.json before sharing workspace configs — it contains endpoint URLs and auth tokens for workspace APIs

FAQ

Q: Is Superset free to use? A: The repository is Apache-2.0 licensed and the core desktop app is free. Some features may require paid API keys for the agents themselves (Claude Code Pro, Codex, etc.).

Q: Can I use Superset with locally running models? A: Superset orchestrates external AI agents via their SDKs. If an agent like Ollama exposes a compatible CLI or MCP interface it can be integrated, but Superset does not natively run local models.

Q: What happens if two agents modify the same file? A: Each agent runs in its own git worktree on different branches. Direct conflicts are rare during the agent phase. Conflicts surface during the merge step — the review panel helps identify and resolve them before merging to main.

Q: Does Superset replace my editor? A: No — Superset focuses on agent orchestration and a terminal view. You still open files in your preferred IDE for direct editing, but Superset sidebar can launch files into any configured IDE.

Q: How is this different from running multiple terminal tabs? A: Multiple tabs run multiple processes but no coordination. Superset provides unified change tracking, automated port forwarding, shared context, and a structured review workflow — each agent runs in an isolated worktree so git history stays clean.

Q: What minimum hardware do I need? A: Superset overhead is modest, but the agents are resource-heavy. A MacBook with 16GB+ RAM is recommended for running 5 or more agents simultaneously without swapping.

Conclusion

Superset is a purpose-built orchestration layer for AI coding agents. Rather than hopping between tabs or manually managing agent outputs, you get a unified TUI where multiple agents operate in parallel worktrees, with shared port routing, MCP integration, and a structured review pipeline. For teams or power users who want to parallelize code tasks across multiple AI models at once, it addresses a real gap between “one agent at a time” terminals and full CI/CD agent pipelines.

If you run multiple coding agents regularly — code review, refactors, tests, docs generation — Superset’s automation model and workspace isolation make it worth exploring. The Apache-2.0 license means you can study the code, fork it, and extend it for custom workflow needs.

Official links:

• GitHub: superset-sh/superset
• Site: superset.sh