Skip to main content
Use an AI chat when you want to think, write, research, analyze a file, or solve a problem with a model. The conversation keeps its earlier messages, so you can refine the answer instead of repeating the full request.

​
Start a chat

1

Select New chat

Use the left rail, or select the + beside the workspace selector and choose New Chat. You can also press Command + Shift + O on Mac or Ctrl + Shift + O on Windows and Linux.
2

Choose a model

Select the model name beside the message box. The suggested model is a good default.
3

Add your request

Explain the result you want, add useful background, and attach files or images when needed.
4

Continue the work

Ask follow-up questions, request a different format, or change models while keeping the conversation context.

​
Give AI better context

A useful prompt normally includes:
  1. the goal;
  2. the background the model needs;
  3. any limits or rules;
  4. the format you want back.
For example:
Turn the attached meeting notes into a one-page project update. Use plain language. Include decisions, open questions, owners, and due dates. Do not invent missing information.
Workspace instructions and relevant workspace knowledge can also help the model answer in your team’s context.

​
Use files, images, voice, and the web

  • Files and images: attach material you want the model to read or discuss. What a model can read depends on that model’s capabilities.
  • Voice input: use dictation to speak your message, then review the text before sending.
  • Web search: turn it on when the answer needs current public information. Answers can include links to sources.
  • Reasoning controls: when available, choose how much effort the model should spend. More effort may take longer and use more credits.

​
Change an answer

You can stop a response, edit your message, or ask for a new answer. Be clear about what should change: Make it half as long, Show the evidence for each conclusion, or Try a more cautious recommendation is more useful than Try again. Changing the model does not require a new chat. This is helpful when you want a second approach with the same context.

​
Know when to start a new chat

Earlier messages help AI understand follow-up questions, but more history can also mean more input for the model to read. As a chat grows, each new request can therefore use more credits, even when some of the repeated input is cached. Start a new chat when you move to a different tender, contract, client, or topic. Keep the current chat when the next question genuinely depends on what came before. A practical pattern is to use an efficient model for routine extraction and a stronger model for the final review. You can see a model’s relative credit-use level in the model picker.

​
Work on a chat together

Chats in a team workspace let members with workspace access see the conversation and continue it. This makes the model’s context and earlier answers available to the team. For work that should stay personal, use a personal workspace. For a topic meant for an active team discussion, use a channel and mention a model or agent there.

​
Share a read-only copy

Use the Share button in the chat header to create a public link. You can share a fixed copy or let the link show new messages as the chat changes. Anyone with the link may be able to read it, so check the full conversation first. Learn how to review and revoke links in Sharing outside Mantle.
Model availability, web search, file limits, and credit use depend on your plan and the selected model. Mantle shows available choices in the model picker.