Skip to main content
New in v3.1. Requires a Coding Plan on the dashboard side — without one, /dashboard/code shows a short upsell in place of this instead of hiding the page.
Turn it on and your Studio session heartbeats to ai.tomdacat.com/dashboard/code — connection status and a live transcript of what the agent is doing, visible from anywhere you’re logged in, not just at the machine running Studio. It’s not read-only: you can send a message into the session from the dashboard too, and optionally let the dashboard approve tool calls right alongside Studio itself.

​
Turning it on

A Remote Control section in Settings, off by default — a plain Off/On toggle, same segmented-control pattern as Budget Mode. The /remote-control slash command jumps straight to it. A live connection-status dot sits next to the toggle, and a second copy of it lives in the chat header itself (beside the credits readout) so you don’t have to leave the conversation to check it:
  • Green — connected, heartbeating normally.
  • Yellow — connecting (the brief window between turning it on and the session actually registering).
  • Red — disconnected.
Clicking either dot jumps to Settings. Both are hidden entirely while the toggle is off, rather than sitting there red for a feature you’re not using.

​
What shows up on the dashboard

ai.tomdacat.com/dashboard/code lists every session currently reporting in, each with a connection dot and a last-seen time. Opening one shows a live (polled, roughly every few seconds) transcript: your messages, the agent’s replies, and a collapsible trace of tool calls grouped together rather than one card per call.
A session only counts as connected based on how recently it last heartbeated, not a status flag that could go stale — closing Studio (or turning Remote Control off) always reflects accurately within a few seconds, it never has to wait for a cleanup job to notice.

​
Sending a message

The composer on a session’s dashboard page queues a plain text message, delivered the next time the plugin heartbeats (a few seconds, not instant — this is polling, not a live socket). It behaves exactly like typing it into the chat box in Studio: it becomes a real turn, the agent responds, tool calls run through the normal local permission gate. Only one message can be in flight at a time — the composer disables itself with “Waiting for Studio to pick up the last message…” until it’s delivered, so you can’t queue up several and lose track of ordering.

​
Remote tool approval

This is a genuine security-posture change, not a default — it has to be turned on explicitly (see below), separately from Remote Control itself.
Normally, the permission gate only ever asks inside Studio. With remote approval on, a mutating tool call waiting on Ask also shows up on the dashboard as a Deny / Always Allow / Allow card — including a real before/after diff for edit_script, the same one the local diff widget would show. Whichever side decides first wins; the other becomes a safe no-op instead of double-firing. While this is on, dashboard access is functionally equivalent to Studio access for approving code execution — anyone who can reach your dashboard can approve execute_luau. Weigh that the same way you’d weigh sharing your Studio session itself.

​
What’s deliberately not here

  • Remote Control only ever sends the settled version of things — a full reply once it’s done, a tool call once it’s started and once it’s finished — not the live token-by-token stream. The dashboard isn’t meant to be a second copy of the chat UI, just a faithful, slightly-behind view of it.
  • Turning off Remote Control (or closing Studio) stops the session from accepting new messages or approvals immediately; anything already queued when that happens is simply never delivered rather than applied later.