AI assistant
Ask questions about your costs, resources and policies in plain language, keep your conversations, and choose the model that answers.
The AI assistant is a chat that lives beside every page. Ask it about your cloud costs, forecasts, budgets, discovered resources and compliance policies in plain language and get an answer without opening each module. It works with your own access: it can only see what you can see in the console.
Looking for something more specific?
| I want to... | Go to |
|---|---|
| Run a task in one step, such as creating a budget | Quick actions |
| Build a dashboard from a sentence | Create dashboards with AI |
| Draft a compliance policy | Draft policies with AI |
| Draft a workspace or generate Terraform | Workspaces and Terraform with AI |
| Connect an AI provider (administrators) | AI provider setup |
Open the assistant
Click Ask AI Assistant in the header. The chat opens on the right and the page you are on stays visible beside it, so you can keep working. Click × in the chat header to close it. Every signed-in user can open the assistant.
In the chat header you also have:
| Button | What it does |
|---|---|
| New Chat (+) | Starts a fresh conversation |
| History | Opens the Conversations list |
| Fullscreen / Exit Fullscreen | Expands the chat to fill the page, or shrinks it back |
| Back arrow | Returns from the conversation list to the chat |
Ask a question
Type a question in the message box and send it. The answer is written as it arrives. While the assistant works you can see what it is doing, for example looking up cost data, and you can click the stop button to interrupt it. Answers use tables and lists where that makes figures easier to read.
The assistant can answer questions about:
| Area | Examples |
|---|---|
| FinOps | Cost over a period, cost by provider or account, forecasts and your budgets |
| Discovery | Which cloud resources you have and how many, filtered by provider, type or region |
| Governance | Which policies exist and how compliant your resources are |
The assistant covers FinOps, Discovery and Governance data. SecOps, Observability and Orchestration data are not available to it, so use those modules directly for answers about them.
Try questions such as:
- "What was our total cloud cost last month?"
- "How many resources do we have in each region?"
- "List the policies that apply to S3 buckets."
Think before answering
Below the message box you can pick a model and turn Thinking on or off.
- The model list shows the models your administrator has connected, grouped by provider. Switching models applies to your next question; earlier answers stay as they were. See AI provider setup.
- With Thinking on (the default), answers that involved reasoning include a collapsible Reasoning Process you can open to see how the assistant got there. Turn it off for a shorter answer with just the result.
Your model and Thinking choices are kept while you work and reset when you reload the page.
Conversation history
Every chat is saved to your account, so it is there after you reload or sign in from another browser. Click History to see your Conversations, each with a title taken from how it started and when it last changed. Open one to continue it, rename it with the pencil icon, or delete it with the bin icon. Click New Chat to start over.
Conversations are private. You only ever see your own, and other people, including administrators, do not see yours in their History.
Long conversations keep working: when a conversation gets very long, the assistant summarizes the oldest turns and keeps the most recent ones in full.
What the assistant can see
The assistant signs in as you for every question. It asks MCM for data with your own access, so it follows the same rules as the rest of the console:
- It can only see the enterprises, accounts and modules your role allows. Two people asking the same question can get different answers. See Roles and access.
- It narrows its answers to the enterprises you have selected in the Enterprise picker. With nothing selected it covers every enterprise you can access. See Console tour.
- If your session ends or your access is withdrawn, the assistant stops working for you straight away.
The assistant is also told to keep its instructions private and not to accept claims about being an administrator as a reason to show more than you are allowed to see.
Questions it asks you
If your request is missing a decision that changes the result, such as which cloud components a workspace should include, the assistant asks you instead of guessing. The questions appear in place of the message box, one at a time, with options you can pick. Some questions let you choose more than one option, and you can always type your own answer. When you have answered them all, the assistant carries on. You can also dismiss the questions and type a message instead.
Tips for good answers
- Be specific about the time range, provider or account you care about.
- Start a New Chat when you change topic, so earlier questions do not get in the way.
- Always check an AI-generated policy, workspace or file before you use it. The assistant drafts; you decide.
"No AI provider configured"
If the footer of the chat reads No AI provider configured, no AI provider has been connected yet and sending is disabled. Hover the notice for the next step: ask a platform administrator to set one up under Settings → AI Assistant Setup. Administrators can find the steps in AI provider setup.