Get startedWork with your agent⌄
Work with your agent
The agent rail runs the agent you already use: Claude Code, Codex or pi. spool ships no agent of its own, and stores no credentials.
Install an agent
Install at least one and sign in to it in a terminal, the way you normally would. spool finds each agent by its command on PATH. With none installed, the rail shows these install lines with copy buttons, and looks again when you press Check again or come back to the window.
- Claude Code
npm i -g @anthropic-ai/claude-code, thenclaude auth login- Codex, 0.151.0 or later
npm i -g @openai/codex, thencodex login- pi
npm i -g @earendil-works/pi-coding-agent, thenpiand/login
The agent menu picks the agent a new chat starts on, for every project on this machine. Threads already started keep their agent and their history.
Logins and models
Each agent uses its own login, models and settings, and spool never asks for a key. The model menu lists what the agent reports on this machine. For pi, models served from this computer, such as Ollama or LM Studio, are marked local.
Permissions
The permission menu in the rail's footer applies to every project on this machine. Its three modes:
- Ask first
- Asks before it edits outside design/ or runs commands outside its sandbox.
- Auto-edit
- Edits files without asking. Asks before commands outside its sandbox.
- Full access
- Never asks.
A command the agent runs inside its own sandbox runs without asking in both asking modes, and edits to design files never ask. pi never asks before it acts, so a pi chat has no permission menu; spool asks pi to say what it is about to change outside design/ first. Stop cancels the work in progress. Closing the canvas leaves a turn running.
Designers
Every agent gets a designer from spool. Ask for options or several directions and the agent gives each one to its own designer with a brief, and the rail shows one row per designer with what it is doing now. A designer runs on the same agent, model and effort as the chat that called it, and is never written into your project or your agent's config.
Your threads
Threads continue in the agent's own saved session. If the agent or its session is missing, the thread stays readable: install or sign in to the agent and press Check again, or start a new thread.