Getting started
From zero to a running multi-agent session in five minutes.
Prerequisites
You need:
- macOS or Linux (x86_64 or arm64)
- Claude Code installed and authenticated
- Git
That's it. initech is a single binary. No Go toolchain, no Node, no Docker.
$ curl -fsSL https://initech.sh/install.sh | bash
On macOS this installs via Homebrew. On Linux it downloads the binary to ~/.local/bin.
Alternatively:
$ brew tap nmelo/tap && brew install initech
$ initech doctor Prerequisites ok initech v1.3.0 ok claude (Claude Code) ok git 2.44.0 ok Terminal: 212x56 (meets 80x24 minimum) All checks passed.
initech doctor checks that all prerequisites are installed and your terminal is large enough. Fix anything it flags before continuing.
$ mkdir myproject && cd myproject $ initech init Project name [myproject]: Source repo URL (blank to skip): github.com:you/myproject.git Select agents for myproject COORDINATORS [x] super ENGINEERS [x] eng1 [x] eng2 [ ] eng3 QUALITY [x] qa1 [ ] qa2 PRODUCT [x] pm [ ] pmm SHIPPING [x] shipper SPECIALISTS [ ] arch [ ] sec [ ] writer [ ] ops [ ] growth Created: initech.yaml Created: 6 agent workspaces (super, eng1, eng2, qa1, pm, shipper) Created: docs/ (prd.md, spec.md, systemdesign.md, roadmap.md) Initialized git repository
The interactive selector lets you pick which agents to run. Use arrow keys to navigate, space to toggle, enter to confirm. Start with the defaults if you're unsure.
This creates:
initech.yamlwith your project config- A workspace directory per agent, each with a CLAUDE.md tailored to the role
- Git submodules for agents that need source access (eng, qa, shipper)
docs/with project document templates
$ initech
This opens the TUI. Every agent gets its own pane in a grid layout. The grid auto-sizes based on how many agents you selected. Each pane runs Claude Code with the role's CLAUDE.md loaded.
You'll see pane title bars with agent names and activity dots: green = agent is working, gray = idle.
Navigate between panes with Alt+Left / Alt+Right. The focused pane has a blue title bar.
$ initech send eng1 "review the README and fix any typos"
This injects text directly into eng1's terminal and presses Enter. The message is delivered synchronously with a guarantee: if initech send returns, eng1 received it.
You can send from any terminal, not just the TUI. Open a separate terminal window and run the command above while the TUI is running.
$ initech peek eng1 -n 20 I'll review the README now and fix any typos I find. Let me start by reading the file... [reads README.md] ...
initech peek reads an agent's terminal output without sending anything. The -n 20 flag limits output to the last 20 lines.
To check all agents at once:
$ initech patrol
$ initech status AGENT STATE BEAD super idle eng1 active eng2 idle qa1 idle pm idle shipper idle
A quick table showing each agent's activity state. No need to peek into individual panes to see who's working.
$ initech down
Gracefully stops all agents and closes the TUI. Agent conversation state is saved in .claude/ directories, so the next time you run initech, agents can resume where they left off.
You can also quit from inside the TUI with Alt+q or the command modal: press ` then type quit.
What's next
You have a running session. From here:
- Configuration to customize your initech.yaml (roles, permissions, agent commands)
- Commands for the full CLI reference
- TUI for all keybindings, layouts, and the command modal
- Roles to understand what each agent type does and how to customize them
- Communication for agent-to-agent messaging patterns
Last updated: April 2026