Lucie — your coding agents, on playbooks

Everything is a file

Playbooks, launchers, watchers — plain TOML in your repo, or global in ~/.lucie. Diff it, commit it, dotfile it. The app is just a nicer editor.

playbooks — the steps

launchers — what starts them

watchers — what guards a worktree

your-repo/

└─ .lucie/

├─ playbooks/

├─ issue-to-pr.lucie.toml

├─ fix-ci.lucie.toml

└─ rebase.lucie.toml

├─ automations.toml

└─ project.toml

~/.lucie/ — global, in your dotfiles

├─ playbooks/

└─ automations.toml

versioned like your code ·
repo files win over globals

issue-to-pr.lucie.toml saved

[[step]]
id       = "plan"
model    = "claude-sonnet-4-6"
pause    = true   # ⚑ waits for your approval
artifact = "02-plan.md"

[[step]]
id       = "implement"
model    = "gpt-5.6-codex"
verify   = "pnpm test"
inputs   = ["02-plan.md"]

Native to your Mac

A real Mac app sitting next to your repos — your worktrees, sessions, and service stacks live on your machine.

Your subscriptions

Runs first-party Claude Code and Codex under your existing plans. No markup, no proxy keys.

Worktree isolation

Each run gets its own git worktree and its own service stack. Agents never trample each other.

Everything on MCP

Every action is an MCP tool — run a playbook, start a terminal, check PR statuses. Drive Lucie from any agent.

Your coding agents, on playbooks

Everything is a file

Put your agents to work