moto is Floom's terminal IDE for AI agents: Claude as the control plane, Codex for backend/infra/debugging, opencode as an optional worker, subagent workflows, Docker-first runtime, CLAUDE.md / AGENTS.md-style context, safety hooks, 60+ skills, Gmail/WhatsApp integrations, memory, browser automation, and cost tracking.

Formerly buildingopen/moto. Old GitHub URLs redirect.

A safety hook firing on a destructive command.

• Founders/operators turning repeated work into software with agents
• Claude Code / Codex users who want a real operating environment instead of one-off prompts
• Terminal-first builders running multiple agents on a Mac + remote Linux box
• Teams standardizing agent rules, hooks, skills, memory, and remote runtime
• Anyone burned by agents running destructive commands, leaking secrets, or claiming success without fresh evidence

Most AI coding setups are a prompt file plus a pile of tools. moto is the full operating environment: context, hooks, skills, memory, terminal sessions, remote server runtime, browser automation, Docker sandboxes, and launcher commands for Claude, Codex, and opencode.

The philosophy is simple:

• context before prompting
• wireframes before UI code
• Docker before SaaS integrations
• subagents before monolithic chats
• cheap models for mechanical checks
• expensive models for judgment
• complexity only after real usage earns it

git clone https://github.com/floomhq/moto.git
cd moto
./install.sh          # Symlinks configs into ~/.claude/

That's it. The installer detects existing configs, backs them up, resolves $HOME paths in hook commands, and installs 17 hooks + 7 scripts + 60+ skills + memory template. Run ./install.sh --copy for standalone files instead of symlinks.

For the full remote workstation from the same repo:

cp .env.example .env
$EDITOR .env
./install.sh mac
./install.sh server-remote

After install:

1. Edit ~/.claude/CLAUDE.md to match your workflow (search for <!-- Customize --> comments)
2. Copy claude/CLAUDE-project.md into your project roots
3. Optionally copy .env.example to .env for API keys used by some hooks/scripts

moto splits agent work between a Mac control plane and a remote Linux runtime. The Mac stays where the human works, while the server owns long-running sessions, containers, cheap sidecar calls, browser automation, and messaging integrations.

flowchart LR
  subgraph Mac["Mac control plane"]
    IDE["IDE + terminal"]
    Claude["Claude Code"]
    Codex["Codex backend/debug"]
    Browser["Local browser"]
  end

  subgraph Link["Connection layer"]
    SSHFS["SSHFS"]
    Systemd["systemd"]
    Tunnels["SSH tunnels"]
  end

  subgraph Linux["Remote Linux runtime"]
    Tmux["tmux sessions"]
    Docker["Docker containers"]
    CDP["Chrome CDP"]
    Messaging["WhatsApp + Gmail"]
  end

  subgraph Surfaces["Agent surfaces"]
    Skills["60+ skills"]
    Hooks["17 safety hooks"]
    Memory["MEMORY.md"]
    Slash["slash commands"]
  end

  subgraph Sidecars["Stateless sidecars"]
    Groq["Groq"]
    Nvidia["NVIDIA"]
    Gemini["Gemini"]
  end

  IDE --> Claude
  Claude --> Codex
  Browser --> Claude
  Claude --> SSHFS
  Claude --> Systemd
  Claude --> Tunnels
  SSHFS --> Tmux
  Systemd --> Tmux
  Tunnels --> Tmux
  Tmux --> Docker
  Tmux --> CDP
  Tmux --> Messaging
  Claude --> Skills
  Hooks --> Claude
  Memory --> Claude
  Slash --> Claude
  Skills --> Tmux
  Tmux --> Groq
  Tmux --> Nvidia
  Tmux --> Gemini

Read the deep dive: docs/architecture.md

Do I need a remote server? No. ./install.sh works locally. Add ./install.sh mac + ./install.sh server-remote only when you want the integrated remote workflow.

Does this replace Claude Code? No. It's the operating environment around Claude Code (and Codex, and opencode). You still use Anthropic's CLI; moto just makes it reliable, safe, and reproducible.

Is this safe? 17 hooks run on every tool call blocking rm -rf /, secret leaks, wrong package managers, unverified WhatsApp sends, and more. See claude/hooks/README.md.

How much does it cost? The repo is free. Hooks track your Claude/Codex token spend to costs.jsonl. Sidecars route bounded work to free/cheap models so you burn premium tokens on judgment, not mechanical tasks.

Can I use only the hooks? Or only the skills? Yes. The installer is modular. Pick what you need.

After running install.sh:

1. Edit CLAUDE.md — Search for <!-- Customize --> comments and replace placeholders with your setup
2. Add project configs — Copy claude/CLAUDE-project.md to each project root
3. API keys (optional) — Copy .env.example to .env for hooks/scripts that need external APIs

The settings.json file has $HOME paths pre-resolved by the installer. Re-run ./install.sh to pick up changes.

Other open-source tools from BuildingOpen:

See CHANGELOG.md.

MIT