Use the AI assistant in a room

What the assistant is

The AI assistant (your "Room Copilot") lives inside each room. Ask it about the room's content, such as PDFs, images, docs, comments, and what was said in meetings, or ask it to do things on the canvas for you.

It can:

• Answer questions and summarize a PDF, a Pactto Doc, a web link, or the room.
• Describe an image or read the text in it, and generate or change images.
• Create and edit Pactto Docs, tables, sticky notes, text, shapes, arrows, and boards.
• Move, resize, align, distribute, or delete items, and edit their text.
• Add comments and set review statuses.
• Run saved instructions called skills. See Use skills in the AI assistant.
• Read from and write to apps you connect, such as Linear. See Connect apps to the AI assistant.

Ask the assistant

1. In the room, click AI Assistant (sparkle icon) in the row of buttons at the bottom right. On a computer the panel is open when you enter the room.
2. Type in Type your instructions or / for skills and press Enter (Shift+Enter for a new line). Or click Dictate a message, speak, and click Stop dictation.
3. While it works, a step line shows what it is doing, such as Thinking or reading the transcript. Click Stop to cancel.
4. Under an answer, use Copy answer, Create doc with this content (turns it into a Pactto Doc on the canvas), or Retry.

Opening the assistant closes the Chat and Live captions panels. Your conversations are saved per room: see Find, rename, and delete assistant chats.

Add files and items to your message

Click + (Add) next to the text box. Under Assets you can:

• Select from canvas, then click items in the room.
• Upload from device. The file goes onto the canvas and is attached.
• Pick a connected source, such as Google Drive, or Add new source.

You can also select items on the canvas first. Attached items show as small tiles above the text box; click the × on a tile to leave that item out. When items are attached, actions apply to those items. The Apps section of the same menu connects outside apps: see Connect apps to the AI assistant.

Ask about a specific item

1. Select a video, PDF, or board on the canvas.
2. Click Ask AI about this asset in the item's title bar. The assistant opens with the item attached.
3. Type your question, or click one of the suggested prompts above the text box, for example Summarize this PDF or What are the action items? (for a Pactto Doc).

On an image, the AI button opens the image AI bar instead, where you can describe a change or use tools such as Erase and Remove background. See Edit images with AI.

Retry or edit your last message

Hover over your last message (on a touch screen the buttons are always visible) to see its time, Retry, Edit, and Copy.

• Retry sends the same message again and replaces the previous answer.
• Edit puts the text back in the text box with Editing message above it. Change it and send; the new answer replaces the old one. Click the X or press Esc to cancel the edit.

Only the last message you typed has these options. Earlier messages show their time and Copy. Anything the replaced answer already put on the canvas stays there.

Questions, suggestions, and results in the chat

• Questions from the assistant: when a request is unclear, the assistant can ask up to five questions in a card above the text box, one per page. Click an option (or press its number key), type your own answer in Other..., or click Skip. Use ‹ and › to move between pages; after the last one, all answers are sent together. Closing the card with X skips the questions that have no answer. Typing in the text box while the card is open answers the current question.
• Suggested next steps: after the latest answer, the assistant may show chips with next steps. Click one to send it as your next message.
• Results on the canvas: when an answer mentions something it created or found on the canvas, the name shows as a chip with a small preview. Hover or click it for a larger preview, then click the preview to select the item and move the canvas to it. A faded chip means the item is no longer on the canvas.

Generate images

When you ask for an image, the assistant fills in a card above the text box called Here's what will be generated before it spends any credits:

1. Check or edit the Prompt. The assistant writes a detailed prompt from your request; whatever is in the field when you confirm is what runs.
2. Pick the Resolution and Ratio, and use One variation more or One variation less to choose how many versions you want (up to 4). The cost updates as you change them. On the Free plan, resolutions above 1K show a Pro badge.
3. If the request makes several images, the card has one page per image (‹ and ›). Settings under All images apply to every page.
4. Click Confirm, or Cancel to stop without spending credits.

While an image is on its way, a small Assistant marker on the canvas shows where it will land, with Waiting for your approval or Generating image. Only you see it. Click it to jump to the card in the panel.

Each finished version shows in the panel as a take. Click Add to room to put it on the canvas. For an edit of an existing image, choose Replace image or Add as new. In the room marks a take that is already on the canvas, and you can add it again without paying twice.

To skip the card, open Assistant settings (sliders icon in the header) and set Confirmation to Auto. Images then go straight to the canvas, one at a time. Ask brings the card back.

Note capturing and proactive suggestions

• Note capturing: when on, the assistant reads the room's messages, discussions, comments, and asset updates to build context for summaries, approvals, and action items. The panel then shows AI context starts from and a time. It turns on by itself once two or more people are in the room, unless someone turned it off. You can pause it any time; the assistant still answers from what it already captured. While it is off, the room's transcript is not saved either, though live captions still show: see Read the transcript and live captions of a meeting.
• Proactive suggestions: when on, the assistant follows the conversation and suggests next steps, like generating or editing assets, summarizing discussions, creating action items, and flagging approvals. Suggestions show as cards with Do it and Cancel; they disappear after 30 seconds if you ignore them. It is off by default. Turn it on or off in Assistant settings.

Only the room owner and people with Can edit access can turn these on or off.

Room instructions

Room instructions tell the assistant how to behave in this room (tone, what to focus on, what to leave alone). They apply to every reply.

1. In the assistant header, click the sliders icon to open Assistant settings, then click Instruction under Agent behaviour.
2. Write your instructions, for example "Summarize in short bullet points. Flag anything that needs legal sign-off."
3. Click Save.

The limit is 65,536 characters. The assistant may also propose changes to the instructions; you can Accept or reject them. Only the owner and people with Can edit access can change room instructions.

Who can do what

• Owner and Can edit: everything.
• Can review: can ask questions, and the assistant can add sticky notes, text, shapes, arrows, lines, comments, and review statuses. Creating or editing other content stops with a message saying edit access is needed.
• Can view: can ask questions, but the assistant cannot add or change anything on the canvas.

If something goes wrong

• The assistant couldn't be reached. Check your connection and try again.: a network problem. Try again.
• Something went wrong: use the retry option on the message, or rephrase the request.
• An image request that the image service turns down because it is busy is retried automatically, up to two times, before you see an error. If it still fails, wait a moment and ask again.
• A permission message: your room role does not allow that change. Ask the room owner for edit access.
• A credits message: the request costs more credits than you have left. See Plans, upgrading and plan limits.

Good to know

• Answers come in the language you write in.
• Most requests cost AI credits. The balance shows in the panel header, and the price of the next request shows under the text box.
• When a reply finishes, the cursor goes back to the text box, so you can type your next message without clicking.
• Changes the assistant makes are made as you and everyone in the room sees them.
• You can also use outside assistants such as Claude or ChatGPT with your rooms through the Pactto MCP server (Connect Claude or ChatGPT to Pactto), or coding agents through the Pactto CLI.