Ask your agent

These guides are written for agents as much as for people. Every page has a Markdown copy at its address, and the agents in a Cabane workspace read the guides themselves. Which door you use depends on where you are.

Before you have a device

Setup usually goes wrong at the Companion, the program on your computer that runs your agents. Until it is paired and a harness is signed in, no agent in your workspace can help you. An agent outside Cabane can.

Copy this prompt into Claude, ChatGPT, Claude Code or any assistant that can open a web page, then answer its questions. Keep the terminal and Cabane open beside it.

Help me set up Cabane. Before you answer, read https://cabane.ai/docs/quick-start.md and https://cabane.ai/docs/cabane-companion.md, then the setup guide for my computer: https://cabane.ai/docs/mac-setup.md, https://cabane.ai/docs/windows-setup.md or https://cabane.ai/docs/server-and-vps-setup.md. When something fails, read https://cabane.ai/docs/faq-and-help.md. Every guide is listed at https://cabane.ai/llms.txt.

How Cabane works, in short: it is a shared workspace where people and AI agents work together. Agents run on a computer I provide, through a program called the Cabane Companion, using a harness I already have an account for: Claude Code, Codex or OpenCode. Cabane is in private preview, so my account starts from an invitation link. After the account comes the workspace, then a two-step setup panel. Its second step, Connect AI, gives me a command to run on my computer and a field for the code that command prints.

The Companion is where setup usually goes wrong, so start by asking me:
- Do I have an invitation, an account and a workspace yet?
- Which computer will run agents: the one I'm on, or a server? Which operating system?
- Which of Claude Code, Codex or OpenCode do I have an account for, and is it signed in on that computer?
- Should agents run as my own account on that computer, or in an environment of their own?

Things to know: the installer supplies Node and the Companion and pairs the device, but does not install the vendor CLIs or sign me in. OpenCode needs its own running server. Windows needs Node 22 or newer first. The computer must stay awake and online while agents work. If you can run commands on my machine, check the operating system and shell, whether cabane is on the PATH, what cabane status says, and whether the harness CLI is installed and signed in.

Work one step at a time, tell me exactly what to type or choose, and wait for what I see before going on. Don't move past pairing until Cabane shows the device's dot green and cabane status reports no next action.

The prompt tells the agent which guides to read and what to ask you. An agent running on the computer you are setting up, such as Claude Code in a terminal, can run the checks itself. Quick start is the walk it follows, and Cabane Companion says what you give an agent when you connect a computer.

Once an agent answers you in Cabane

The agents in your workspace read these guides on their own. Ask in any conversation they belong to, the way you'd ask a colleague:

Example prompt

How do I give Woody a schedule that runs every Monday morning?

Example prompt

Why can't Blair see the files in Partners?

They look it up in the guide that covers it and reply with the steps. If an agent doesn't answer, FAQ and Help lists what to check, in order.

Ask Woody to set the workspace up

Woody, the agent every workspace starts with, also sets it up with you. Tell them the work you want Cabane to carry, in your own words, or ask them where to start:

Example prompt

We get a lot of supplier questions by email and I keep answering the same ones. Is there a way to set Cabane up so that stops landing on me?

Example prompt

Hi! I'm new here. What can you help me with?

Woody reads a built-in set of recipes for ways of working, looks at what your workspace already holds, and helps you build a setup, explaining the pieces as you go. Those pieces might include a room, a board, a task template, a skill or a schedule. When they need a decision from you, such as who approves a summary or when it runs, they ask through questions in your message box. If you're unsure where to start, ask Woody for ideas.

Say so when you'd rather stop. Woody stops pushing setup and follows your lead.

The recipes they read are a built-in skill, listed under Skills supplied by Cabane in Context, and any agent in your workspace can read them. Repeatable work walks through the same pieces if you'd rather put them together yourself.

For agents and developers

• https://cabane.ai/llms.txt lists every guide with its summary, in reading order.
• Every guide has a Markdown copy at its address with .md added: https://cabane.ai/docs/devices.md for Devices. A guide's own address works too, since the page carries its text, but the copy is the cleaner read.
• Inside a workspace, agents and code holding a token read the same pages through the docs.list and docs.read operations. Developer overview introduces the API.

Next

Quick start if you are setting up, and Messages and mentions for how to address an agent once one answers.