--

Every coding-agent CLI in one place. A desktop app (Tauri + xterm.js): add a project folder, then spawn Claude Code, Codex, Cursor, or a plain terminal in any directory - tiled side-by-side, each with a live activity dot and a workflow status you set.

Download the latest .dmg from Releases, open it, and drag tt to Applications.

The build is unsigned, so on first launch: right-click tt -> Open -> Open (only needed once). Then install the agent CLIs you want (see Requirements).

• Projects - add a folder via the native picker; switch via the top-bar dropdown (persisted).
• Directory tree (right panel) - browse, create folders, and open an agent in any folder.
• Auto-grid tiling of all running agents; click a tile to zoom (focus mode).
• Status - a live dot (working / idle / exited) plus a workflow tag you set: Planning · In progress · In review · Done (it colors the tile).
• Rename agents (double-click the title); collapse the left/right panels.
• × kill an agent from its tile or the rail.

• Rust (rustup) and Node 20+

• tmux (optional) - agents run inside it so their sessions survive closing/restarting the app; without tmux they still work but don't persist.

• The agent CLIs you want to use, on your PATH:

  • claude - Claude Code
  • codex - Codex CLI
  • cursor-agent - Cursor CLI
  • gemini - Gemini CLI
  • opencode - opencode CLI
  • antigravity - Antigravity CLI
  • a shell for the plain terminal (uses $SHELL)

  Gemini's free Google sign-in was deprecated for the CLI. Set a GEMINI_API_KEY (free key at aistudio.google.com/apikey) or choose "Use Gemini API Key" in its prompt.

# from the repo root
source scripts/dev-env.sh   # puts cargo + node on PATH (see note below)
npm install
npm run tauri dev           # compiles the Rust, opens the app window

Production build: npm run tauri build.

scripts/dev-env.sh exists because this machine loads node through an nvm lazy-loader that only works in interactive shells; it drops the stub and points PATH at node + ~/.cargo/bin. If cargo and node are already on your PATH, you can skip sourcing it.

Each agent is launched in the folder you pick:

Claude gets a pinned --session-id so tt reads exactly that session's title + token count from ~/.claude/projects (no cross-talk between two claudes in one folder).

While tt is running it exposes an MCP server at http://127.0.0.1:4127/mcp (loopback only, no auth). Point any agent CLI at it and that agent can spawn and coordinate more agents — the new agents appear as normal tiles in tt.

Tools: spawn_agent(agent, dir) · list_agents() · send(agent_number, text) · broadcast(text, numbered?) · close_agent(agent_number). Agent numbers are hierarchical — a sub-agent spawned by #1 is 1-1.

Add it to Claude Code (restart the session afterward):

claude mcp add tt --transport http http://127.0.0.1:4127/mcp
# verify:  claude mcp list   →   tt … ✔ Connected

Every other CLI uses the same URL — see docs/MCP.md for Codex, Cursor, Gemini, and opencode config snippets.

source scripts/dev-env.sh
cd src-tauri && cargo test    # Rust: registry, PTY, jsonl parsing, fs
cd .. && npm test             # frontend: store + tiling math

docs/superpowers/specs/ (design) and docs/superpowers/plans/ (implementation plan).