Your workspace, from the terminal or the AI you already use

Use the AI you already use

utilityspaces connect adds the workspace to Claude Code, Codex or opencode as a set of tools. Any other program that speaks the Model Context Protocol — GitHub Copilot, Cursor, Cline — gets the same tools when you add utilityspaces mcp to it as a stdio server. The program stays what it was: your subscription, your model, your conversation. What it gains is the ability to list members, create an agent, open a task or read what changed today, in the workspace you signed in to.

Reads and everyday changes run as they would for you in the web app. A high-risk change, such as removing a member, comes back as a plan with a token, and nothing applies until the program sends that token back. The program holds that token, so whether you see the plan first depends on the program, not on UtilitySpaces: keep its approval prompt on for high-risk tools, or give it a read-only sign-in. utilityspaces login --scope read makes a sign-in that can look at everything you can and change nothing.

Or chat in the terminal

Typed on its own, utilityspaces opens a chat: sign in by approving a short code in the browser, pick a brain, then say what you want done in plain words. The brain is yours to choose: an AI coding tool already on your computer (Claude Code, Codex, opencode, Kilo Code or GitHub Copilot, each on your own account), a key of your own, a model server on your computer such as Ollama, LM Studio, llama.cpp, vLLM or Jan, or the workspace's own model.

Where your workspace's data goes depends on that choice, and the chat says so when you make it. With a model server on your computer, your messages and what the workspace returns stay on your computer. With a key of your own they go to that key's provider. With an AI coding tool they go through it to whichever provider runs the model it uses, on that provider's terms. With the workspace's own model they go to UtilitySpaces's servers, and on to the provider of a key the workspace saved if it uses one. A workspace holds people's names and emails, so this is the fact to pick by. With every choice but the last, UtilitySpaces receives the operations the chat runs and never the conversation around them.

An AI coding tool in the chat is started with the workspace's tools and nothing else: no shell, no file edits, no web. If one starts a tool of its own anyway, the chat ends the turn and says which tool it was. Every call shows as a row with its time. A high-risk plan appears as a card and waits for your key, and things shown once, such as an invitation link, go to your screen and not to the model.

What it does today

More than 120 operations: members and invitations, agents and the brains they run on, channels, tasks, schedules and flows, CRM records, data tables, skills, knowledge, connectors and ready-made recipes. That is about 25% of what the web app can do, and utilityspaces ops lists exactly which. The rest is in the web app, where it has always been; the command line sits beside it, not in its place.

What waits for you

The command line goes through the same gates as the web app. Your role decides what you may do, on every call. A sign-in covers the one workspace you approved it for, lasts 30 days, and ends sooner with utilityspaces logout or from Settings. Changes are written to the workspace's audit log with who made them. Add --dry-run to a change to see its plan first; a change with no preview says so and does nothing. --json gives output a script can read.

When something fails, utilityspaces doctor says what stands between that computer and a chat, and utilityspaces report tells support about a failed command. What is sent is the reference the server recorded and your note, never the values you typed.

Getting it

There is nothing to download from this page. The command rides in the brain app, because that is the one piece of UtilitySpaces a workspace already keeps on a computer it controls. It updates when the brain app does: Update brain runs the installer again, and a brain app from before the command line needs that once.

1. In a workspace, open Brains → Add brain, pick one, and run the one-line installer it shows. A demo workspace can connect 1 brain of its own, and the command comes with it.
2. Run & "$env:USERPROFILE\.utilityspaces-daemon\runtime\utilityspaces.cmd" login in PowerShell on Windows, or ~/.utilityspaces-daemon/runtime/bin/utilityspaces login on macOS and Linux, and approve the code in your browser.
3. To type just utilityspaces from then on, put the brain app's folder on your PATH: %USERPROFILE%\.utilityspaces-daemon\runtime on Windows, ~/.utilityspaces-daemon/runtime/bin on macOS and Linux.

Sign in to a workspace if you have one, or see what a workspace costs.