Conversations in Studio

A conversation is where you ask Studio for work. Each conversation belongs to a folder (a project) and uses its files.

Start

1. Click New conversation in the sidebar, or Start in this project in a folder’s menu.
2. In the prompt box, describe what you need in your own words and send it.

While typing, use @ to point to a file in the folder and / to call a skill or a connector. Attach adds files and images. If the chosen model only reads text, Studio warns you before you send an image it will not see.

Dictate

Dictate records your speech and turns it into text in the prompt box. Transcription happens on your computer: what you say does not leave the machine. The first time, Studio asks to download a transcription program (Download and install), only once.

Model and access

The prompt box holds two choices for the conversation:

• Model and thinking: the AI provider, the model, and the Reasoning effort (from Minimal to Max, depending on the model). A change of effort applies from the next question on.
• Access: Automatic assesses each action and asks for your approval when needed; Full Access runs without asking for approval. When you choose Full Access, Studio asks for confirmation, and system permissions still apply.

While Studio works

• You can Write the next prompt: it goes into the Prompts waiting their turn queue. In the queue, you can Send now, Edit, or Remove each one.
• Stop interrupts the work in progress.
• The Progress card shows the task list Studio built and what is already done, plus terminals left running in the background.

Approvals and questions

When an action needs your permission, the conversation shows This needs your permission, with Allow, Always allow in this conversation, and Refuse. For commands, Studio explains what the command does (reads only, writes to the folder, downloads from the internet, installs on this computer) and offers Run or Do not run.

When Studio asks a question, answer it in the prompt box.

While the conversation waits for you, it shows as Waiting for you in the list and on the stages board.

Plans

For larger work, Studio may save a plan before executing it. The plan card offers Open plan and Execute plan. When executing, choose:

• where: In this conversation (continue with the current history) or In a new conversation (start clean in the same folder);
• the Execution mode, the provider, and the model;
• Additional instructions, optional, that apply only to that run.

When it finishes

The Work completed card summarizes what was delivered. Sometimes Studio suggests a related task that is outside the request; on the suggestion card, choose New conversation, Do it here, or Dismiss.

If your subscription hits its limit, the conversation tells you when it renews (when the provider says so). You can wait, choose another model, or use another provider.

Go back and branch off

On any of your messages:

• Go back to this point resumes the conversation from there; Go back to this point and revert file changes also undoes the file changes made afterward. If you edited those files later, Studio warns you what you will lose.
• Branch off from here creates a new conversation with the history up to that point, without touching the original.

The conversation side panel

The side panel shows About this conversation (folder, provider, model, thinking, access, memory, and Usage, with credits spent) and opens, in tabs, Files, Terminal, Version control, and Side conversation. In the side conversation you ask about the conversation without changing it: answers only read, and nothing goes into the main conversation.

To find a passage, use Search this conversation.

Organize conversations

In each conversation’s menu: Rename, Pin, Mark as unread, Fork, Archive, and Delete permanently. Deleting removes the conversation’s history and attachments and cannot be undone; files in the working folder are preserved. Archived conversations can be restored.