# What agents can do in Studio

> Public channel — documentation of app 3.6.3. Follow the availability notices of each service.

Source: [What agents can do in Studio](/en/application/studio/agents-dans-studio/)

The Studio tools are provided by VibeStarter. There is nothing to enable in Roblox Studio.

The plugin gives your agents real access to the open place: they can read it, run code in it, start a test and look at the result. This page says how far that goes, what is deliberately refused, and why only one agent drives Studio at a time.

## What an agent knows how to do

| Family | What the agent can do |
| --- | --- |
| **Read** | search the tree of your game, inspect an object and its properties, read the server console and the client console, read and change the selection in Studio |
| **Act** | run code in the place, insert a Roblox asset by its ID |
| **Test** | start a test and stop it, simulate keys, clicks and typing, move a character, simulate a mobile device profile, measure performance, replay a recorded test scenario, with up to eight players |
| **Capture** | take a picture of what Studio displays, or a burst presented as a contact sheet, by framing a target and then putting the camera back where it was |
| **Audit** | report structural defects in a map, the rigging of a model, the quality of an asset, the layout of an interface, and list the assets of the game |

## What is refused to it

-   **An agent does not change scripts from Studio.** It edits the files of your project, and the change then arrives through sync. A second write path would make the place diverge from your code. This is a matter of principle, not a technical limit.
-   **An agent does not link a place to a project.** That step is yours, from the app. See [Link a place to your project](/en/application/studio/lier-une-place/).
-   **An agent does not use your other Studio plugins** (animation tools, effects suites, third-party plugins).
-   **An agent does not publish your game.** See [Publish your game and track your versions](/en/application/studio/publier/).
-   **An agent does not stop a test it did not start.** A test you started by hand, or that another agent started, stays under the control of whoever opened it.

## One agent per place: the queue

An agent reserves the place it works on, so that two agents do not get in each other’s way. The others carry on meanwhile with what does not need Studio: the code, the assets, the thinking.

Two-part diagram. At the top, three agents and one place open in Studio: agent 1 reserves the place, agent 2 waits for a place in position 1, agent 3 carries on with the code and the assets without Studio. At the bottom, two worlds of the same project: agent 1 drives world A and agent 2 drives world B, in parallel, one reservation per place. — One place: one agent at a time — Agent 1 — Agent 2 — Agent 3 — Place open in Studio — Studio busy · Agent 1 — 0/1 places free — reserves the place — Waiting for a place (position 1) — Carries on with code and assets — without using quota while waiting — Two worlds in the project: two agents in parallel — Agent 1 — Agent 2 — World A · busy · Agent 1 — World B · busy · Agent 2 — one reservation per place

One place can only be driven by one agent at a time: the second waits, and the third does not need Studio and carries on. With two worlds in the project, two agents work in parallel, one per place.

In the app, the states are explicit:

| What is shown | What it means |
| --- | --- |
| Studio free | no place is reserved |
| `Studio busy · {{qui}}` | an agent holds the place |
| `{{libres}}/{{total}} places available` | how many places remain available |
| `Waiting for a place (position {{n}})` | your agent is in the queue |
| `Take {{place}} back` / `Stop · {{place}}` | take control back or free the place |

The queue does not waste your quota

An agent waiting for Studio does not use any quota with your AI provider. If you have several worlds in your project, two agents can drive two worlds in parallel: the reservation is per place. See [Several worlds in one game](/en/application/studio/plusieurs-mondes/).

## Game tests

An agent starts a test through the test functions of Roblox Studio, just as you do from the toolbar. Two modes exist, and the difference matters:

-   **A test with a player** starts a full session, with a character. It is the only mode that allows a capture or the inspection of the interface.
-   **A server-only test** starts the server without a client. No capture is possible there, and the agent tells you so instead of waiting.

An agent can also replay a recorded test scenario with your project. The verdict is a pass, a fail, or **inconclusive** when the necessary evidence is missing. An honest result is worth more than a false pass. It can finally run the same scenario on one to three mobile device profiles, with a verdict per profile, then put Studio back in its original state.

Keep the Studio window restored

When the Studio window is minimized, the editor’s rendering is paused: captures fail, and a simulated click on a button may not trigger its action. For any visual check or interface test, keep Studio visible on screen.

See also [Have an agent test a feature](/en/application/agents/playtest/).

## Only one test per game at a time

Roblox Studio plays only one session of a given experience at a time. Two worlds of the same game therefore cannot be tested at the same time. Two different games can.

## What goes to your AI provider

The app says so when you authorize the plugin: “What they consult (content, logs, captures) goes to your agent’s AI provider.” and “The plugin never joins online games and never reads player data.”

In practice, what an agent looks at in your place (the game tree, the console messages, the captures) goes to the AI provider you connected, just like your code. VibeStarter does not connect to your published game.

VibeStarter asks for no system permission for any of this: no screen recording, no accessibility, no bringing the window to the front.

## Limits

-   The available tools depend on the version of your app. A Studio window running an older plugin refuses what it does not know. Restart Studio after an update.
-   A capture is impossible in a server-only test, and in a multiplayer test. For a visual check, run a test with a single player.
-   A minimized Studio window can stop drawing, especially while editing. The capture fails and the tool says so. Restore the window and ask again.
-   Nothing guarantees that a future version of Roblox Studio will keep exactly the same behavior.
-   An agent can be wrong about its own abilities and claim it cannot do something. Ask again explicitly, or start over in a new conversation.

## Frequently asked questions

**Why do my agents wait for Studio?** Because only one agent drives a place at a time. This is normal, and it does not interrupt their work on the rest.

**Will several agents build my map faster?** No. Reserving the place puts them in a queue. To run work in parallel, give them tasks that do not all need Studio.

**The agent says it is not connected to Studio.** Check that the place is open, saved on Roblox and linked to your project. Then restart Studio. The rest is on [the troubleshooting page](/en/aide/studio/).

**My test stays stuck and Studio no longer responds.** Stop the test by hand in Studio, then start it again. If Studio stays frozen, close it and reopen it.

## Going further

-   [What your agents can do](/en/application/agents/ce-qu-ils-savent-faire/)
-   [Have an agent test a feature](/en/application/agents/playtest/)
-   [Launch and drive your agents](/en/application/agents/lancer-et-piloter/)
-   [Synchronization, explained](/en/application/studio/synchronisation/)
-   [The Roblox Studio connection does not work](/en/aide/studio/)