Conversations and permissions
Ask questions, add materials and manage the assistant's actions.
Start and continue a conversation
Choose an assistant, open a new conversation and send a message. Keeping one task in each conversation makes follow-up questions easy to find alongside the original materials and result.
To resume work later, open the conversation from your history. Unsent text in the input field remains a draft: the assistant starts working after you send the message.
Add context
State the goal, source information, constraints and desired result format. Add a file if the task concerns a document or spreadsheet. For larger collections of materials, use a project or assistant knowledge.
If the result is too general, refine a specific part: “Add an example to the third point” or “Keep only the tasks for this week.”
Allow or reject an action
The assistant may request permission to use a tool. Read what it intends to do and which data it will use, then choose the appropriate response.
- Allow once — perform the proposed action.
- Allow for this session — allow matching actions in the current working session without asking again.
- Reject — do not perform the action; explain what should change if necessary.
Permission for a session does not grant access for all future conversations. Additional restrictions set by the owner in the assistant's profile may also apply.
Answer an assistant's question
When more information is needed to continue, respond in the question card. Granting permission and answering a question are different actions: check that you have addressed the request that is waiting.
Conversation status indicators help you find chats that need attention. Open the conversation, respond and check that work continues.
Stop work or recover a response
To stop the current run, use the stop button in the conversation. Stopping does not undo external actions that have already completed, such as sending a message or changing a document.
If the response stops updating, reopen the conversation history. Before sending the request again, check whether the result is already there: losing the browser connection does not always stop the assistant. See troubleshooting for more steps.