kortix-computer

name

kortix-computer

description

How to reach a CONNECTED MACHINE (a user's laptop/desktop, or any computer paired over the Agent Computer Tunnel) from a Kortix session — read/write files, run shell commands, and drive the desktop (click/type/screenshot) on that machine. It works through the `computer` connector and the same list/discover/show/call path as every other connector, so there is no separate tunnel client and no token. Load this when the task is about acting ON a specific physical/remote computer the user has connected ("on my laptop…", "read ~/Downloads on my machine", "run this on my desktop", "click the button on my screen"), or when the user asks how the agent reaches their computer. For files INSIDE this sandbox, just use normal shell/fs — not this.

A user can connect their own machine to Kortix over the **Agent Computer Tunnel** (a permissioned reverse tunnel). Once connected, you reach that machine through the **connector gateway** — it shows up as a single **`computer`** connector that fronts **all** of the account's connected machines. You use the normal `kortix-connectors` MCP tools (`connectors` → `discover` → `describe` → `call`); there is **no token and no separate tunnel CLI** — the live tunnel is the credential, resolved server-side.

The computer connector's tools relay an RPC to the machine:

• filesystem — computer.fs.read / fs.write / fs.list / fs.stat / fs.delete
• shell — computer.shell.exec (stdout / stderr / exitCode)
• desktop — computer.desktop.cua.click / type_text / press_key / hotkey / scroll / launch_app / list_apps / list_windows / get_screen_size / get_accessibility_tree, plus computer.desktop.cua.call (a passthrough to ANY computer-use tool by name).

Every relayed tool takes a computer argument selecting which machine (its name or id). It's optional when exactly one machine is online — then that one is used by default.

This is for a connected, external computer — not this sandbox. To touch files in your own workspace, use normal shell/fs. Reach for computer.* only when the task is explicitly about the user's own machine.

**1. See which machines are connected.** Call the connector `call` tool:

{ "connector": "computer", "action": "list_computers" }
// → { "computers": [ { "id": "…", "name": "Marko's MacBook", "online": true,
//                      "capabilities": ["filesystem","shell","desktop"], "platform": "darwin" } ] }

If connectors doesn't list a computer connector at all, this project cannot reach a machine yet. Say so plainly and hand back the two steps — don't hunt for another way in:

1. Turn the feature on. The Agent Computer Tunnel is gated by the agent_tunnel experimental feature. A member with customize-write enables it in Customize → Settings → Experimental. Until it is on, the Computers section does not appear in the dashboard at all.
2. Pair a machine in Customize → Computers, which shows up once the feature is enabled.

If the connector IS listed but list_computers shows the target online: false, the machine is paired but not running — ask them to bring it online.

2. Call a tool, picking the machine. Pass computer (name or id). Omit it when only one machine is online.

// read a file on the laptop
{ "connector": "computer", "action": "fs.read",
  "args": { "computer": "Marko's MacBook", "path": "/Users/marko/notes.md" } }

// run a command (sole online machine → no selector needed)
{ "connector": "computer", "action": "shell.exec",
  "args": { "command": "git", "args": ["status"], "cwd": "/Users/marko/proj" } }

// drive the desktop
{ "connector": "computer", "action" 
         

describe any tool first if unsure of its inputs (e.g. { "tool": "computer.shell.exec" }).

3. The passthrough for the long tail. Beyond the curated desktop actions, computer.desktop.cua.call invokes any computer-use tool by name:

{ "connector": "computer", "action": "desktop.cua.call",
  "args": { "computer": "Marko's MacBook", "tool": "double_click", "args": { "x": 220, "y": 140 } } }

The machine's owner grants access **per capability** (filesystem / shell / desktop), scoped (allowed paths, allowed commands, allowed desktop features), in **Customize → Computers**. The tunnel layer enforces this on every call — the The connector gateway does not bypass it.

If you call something that isn't yet granted, the call comes back as pending_approval: a permission request is created and surfaced to the user in Computers. Tell them what you're trying to do and that they need to approve the request in Computers, then retry once they have. Don't try to route around a denial (there is no token to fall back to, by design).

- **Use the `computer` connector** — never hand-roll a tunnel client or look for a tunnel token. There isn't one in the sandbox by design. - **Pick the machine deliberately.** When more than one is online, always pass `computer`; a call without it errors and lists the options. Use `list_computers` to choose. - **Be careful with write/destructive ops.** `fs.write`, `fs.delete`, `shell.exec`, and desktop control act on someone's real machine — confirm intent for anything irreversible, exactly as you would locally. - **`pending_approval` is not a failure** — it means "ask the human to grant it in Computers", then retry. Surface the request plainly. - This is for an **external connected machine**. For this sandbox's own files, use normal shell/fs, not `computer.*`.