AI agents

How to connect

1. On the Mac where your agent runs, open “Settings → AI agents” and turn on “Allow AI agents”.
2. Under “Connect to AI”, press “Connect” next to your program: Claude Code, Claude Desktop, Codex, Cursor or OpenClaw. EnotRemote adds itself to its settings and gives the agent a skill with the rules of work. For another program, such as Windsurf, VS Code, Zed, Gemini CLI or JetBrains, there is a line for its MCP settings, and EnotRemote knows it by its mark.
3. Tick “AI access” for the computers the agent may work with: in the settings or in the computer’s menu (the “…” button). The agent doesn’t see the others.
4. The first time it calls, the agent’s program asks for permission: “Allow”, “Only now” or “Deny”. Allowed programs are listed in the settings, where you can revoke them.

What an agent can do

• See the computer’s screen and take screenshots.
• Click and type, including keyboard shortcuts.
• Read and put text on that computer’s clipboard.
• Send files there from allowed folders: the AI program’s project folder, Downloads, Desktop, ~/EnotRemote and the ones you added. Hidden folders never.
• Download files from that computer: they go to Downloads, into the EnotRemote folder.
• See which window is in front, which field has the caret and what is already written in it.
• Get a list of a window’s buttons and fields and press them by number instead of hunting on a screenshot.
• Switch the keyboard layout on that computer: before typing by keys it sets English itself, checks it and then puts the previous one back.
• Work in a tab with that computer you already have open: when the agent is done, the tab stays yours.

You always see it and can stop it

• The tab where the agent works glows softly and shows its program’s mark: Claude, Codex, Cursor and others. “Take back control” stops the agent at once, even in the middle of typing, and “Give back to the AI” hands it back.
• The agent’s cursor carries its name, and blue arcs glow along the edges of the picture. Your mouse movements don’t get in the agent’s way, and your click or key takes back control.
• The other computer shows a pill saying the AI agent is in control, with its name. The person there stops the agent with a button or Ctrl+Alt+Shift+Q, and can forbid agents altogether in the settings: “Let AI agents in”.
• An agent’s session closes by itself after 10 idle minutes.
• The “Activity log” shows everything the agent did.

What an agent won’t do

• Press keys when the wrong window is in front: nothing is sent, and the agent learns which window is in front.
• Type over text that is already in a field by mistake; after typing it checks that the field holds exactly what it typed.
• Get the text of password fields.
• Delete files in Explorer, Finder or on the desktop unless you asked, and it never deletes them for good.
• Open password managers (KeePass, 1Password, Bitwarden), system settings or EnotRemote itself on that computer.

Where it works

For now the agent runs only on a Mac. It can control Mac and Windows computers that have the latest EnotRemote. Windows is coming soon. While in beta, it is free.