COReClaw connects your machines to CORe so you can ask the AI to actually do things on them. Run commands, read and edit files, manage services, check on a process, deploy something, all by describing it in plain language. Pick a server, type what you want, and a CORe model carries it out on that machine and streams the results back.

You add a machine by running one install command on it. It shows up in your fleet a few seconds later, and from then on it has its own persistent chat. Each server keeps its own conversation, so the AI remembers the context of that specific box rather than mixing everything together.

Install a machine

Open COReClaw from the dashboard and click Add server. Give it a name you'll recognize, like "prod-web-1" or "Home PC". You'll get two commands, one for Linux/macOS and one for Windows. Run the right one on the machine you want to connect.

  • Linux / macOS uses curl to fetch the installer and run it. It sets the agent up as a background service that comes back on its own after a reboot.
  • Windows uses a PowerShell command that does the same thing, and adds a Startup entry so it relaunches when you sign in.

Within a few seconds the server flips to Online in your fleet. The token in the command is what links that machine to your account, and it's only shown once, so keep it private. If a machine ever needs to be re-linked, remove it and add it again to get a fresh command.

User mode vs Root mode

During install you're asked one question: install as User or Root. This controls how much the agent is allowed to do on that machine.

  • User (limited access) is the safe default. The agent runs with normal user permissions. It can read and edit files in your home directory, manage user-level services, and do everyday tasks. Choose this for personal machines, anything with sensitive data, or when you're not sure. On Windows this just works. On Linux it runs as your user account.
  • Root (full access) runs the agent with administrator privileges, so it can manage system services, install packages, edit system config, and touch anything on the box. Choose this for servers you fully own and administer, where the whole point is to let the AI operate the machine. On Windows it asks for a UAC prompt. On Linux it asks for your sudo password during install.

You can switch a server's mode later from the COReClaw dashboard. Click the three dots next to the server and toggle the privilege mode. The agent on that machine updates itself, so you don't need to reinstall.

Using a server

Click a server in the sidebar and you're in its chat. Type what you want in plain English and the AI does it. A few things worth knowing:

  • Pick your model. The model selector in the top bar works like the main chat. Heavier models are better for complex multi-step work; the fast ones are fine for quick checks.
  • Watch it work. When the AI runs a command or edits a file, you see it happen live in a tool card. Tap the card to see the exact command and its output.
  • Stop anytime. While it's generating, the send button turns into a Stop button. Click it to halt the response immediately.
  • It remembers. Each server keeps its own history, so you can come back later and it still knows what you were working on. You can also let agents see your main-chat memories from the COReClaw settings (the cog at the bottom of the sidebar) if you want them to personalize responses.

Command it from your phone

There's a COReClaw Android app for managing your fleet away from your desk. It has the same server list, the same per-server chat, and the same controls. Grab it from the downloads page, or from the promo card in the COReClaw dashboard.

Removing a machine

To take a machine off your fleet, open its three-dot menu in the sidebar and choose Uninstall agent. That gives you a one-line command to run on the machine that stops and removes the agent cleanly. Then hit Remove to drop it from your list.

A note on trust

COReClaw is powerful by design: Root mode lets the AI change anything on a machine. Give Root only to machines you own and trust, prefer User mode anywhere you're unsure, and read destructive commands before you ask for them. The agent is built to confirm risky operations in plain text before it runs them, but you're still the one in charge of the machine.

Open COReClaw Get the app