Starting and Managing Conversations
Typing and Submitting a Question
To start a conversation, type your question or prompt in the input field at the bottom of the AI Companion panel or full-page view, then press Enter or click the send button.
There is no special syntax required. Write in natural language the same way you would ask a colleague: "Which devices haven't checked in with any endpoint agent in the last 30 days?" or "Are there any open cases assigned to nobody?" The agent interprets your intent and determines the best way to respond.
If your question is ambiguous or the agent needs more information to build an accurate response, it will ask you a clarifying question before proceeding.
How the Agent Responds
The agent's response streams into the conversation as it is generated, so you see the answer build progressively rather than waiting for the full response to appear at once.
Depending on what you asked, the response may include:
- Text β A plain-language explanation, summary, or answer.
- A table β When the agent returns asset results or structured data, it formats them as a table with clickable links that open the corresponding record or asset page in Axonius.
- Tool execution steps β The agent shows when it is calling a tool (such as querying assets or looking up a saved query) before the result comes back. These steps are visible in the conversation thread so you can see what the agent is doing.
- Suggestion chips β After some responses, the agent appends clickable follow-up prompts. See Suggested Actions.
- Stop generation β While the agent is generating a response, a stop button appears in the input area. Click it to halt generation and keep the partial response returned so far.
Ask AI - Asking About What's on Screen
Use Ask AI to ask the AI Companion about the screen you're on. Axonius adds a ready-made question about the screen to the chat. Review the question, and then send it.
For example, on the Advanced Configuration tab of an adapter, you turn on two settings and set a number setting to 500, without saving. When you select Ask AI, the message box displays a question listing the three changes and asking what they do and whether it's safe to save.
Required Permissions
To use Ask AI, your role needs the AI Agents > View and interact permission. The button doesn't appear for a user who impersonates a data scope. Ask AI appears on every page in Axonius.
Using Ask AI
To ask about a screen:
- Select Ask AI on the screen. A new chat opens next to the page, with the question in the message box.
- (Optional) To change the question, edit the text.
- To send the question, press Enter.
Consider the following:
- Axonius doesn't send the question until you send it, and no AI credit is used until then. To ask your own question, delete the text and type a different question.
- The message box grows to show a long question, up to 12 rows. While the question waits in the message box, the suggestion cards are hidden. They appear again when you clear the message box.
- When you have unsaved changes, the question lists them. A toggle appears as "turned on" or "turned off". A number or a list choice appears with its new value. For free text and lists, the question names only the setting and never includes what you typed.
- The question never includes a password, file, or vault field.
- On a Security Finding (one finding on one asset), Ask about this asset drafts a question about that finding. See Asking About a Security Finding.
- The AI Companion reads the saved values and explains each unsaved change against them.
Asking About a Security Finding
Use Ask about this asset on a Security Finding to draft a question about that one finding on its asset. The permission is the same as for Ask AI: AI Agents > View and interact.
To ask about a Security Finding:
- On the Security Findings page, open a Security Finding. The finding opens on its Security Finding Profile page, or in the asset drawer when you open it from the table.
- Select Ask about this asset. A new chat opens with the question in the message box.
- (Optional) To change the question, edit the text.
- To send the question, press Enter.
The question asks the agent to explain the following:
- What the finding is.
- How attackers exploit it.
- How to remediate or mitigate it on this asset.
- How exposed this asset is: internet-facing, internal, or in a sandbox.
- Which adapters to connect when your data can't tell.
For example, on a finding named "CVE-2099-0001 on host-1", the message box shows:
Explain the security finding "CVE-2099-0001 on host-1" β what it is, how attackers exploit it, and how to remediate or mitigate it on this asset. How exposed is this asset (internet-facing, internal or in a sandbox), and which adapters should I connect if my data cannot tell?
The Security Finding Profile page and the asset drawer show the same question. Axonius doesn't send the question until you send it. An Aggregated Security Finding and an IAVM finding keep their own questions.
When a Request Reaches the Step Limit
One answer of the AI Companion has a fixed step limit. A large request, for example a dashboard with many charts, can reach the limit before the AI Companion finishes. No setting changes the limit.
When the AI Companion reaches the limit, it stops and tells you what it finished. Anything it already created stays in place. To go on from where it stopped, send "continue".
When the AI Companion stops before it can write an answer, the chat shows this message: "I stopped because this request reached the step limit for one answer, before I could write my answer. Nothing is still running, and anything that I already created stays in place. Send "continue" and I will go on from where I stopped."
Agent Routing
When you send a message, the AI Companion system determines which specialist is best suited to handle it and routes the request accordingly. You will see the active agent name reflected in the panel header and, when routing occurs, a brief routing indicator in the conversation thread.
You do not need to address a specific agent or use any special command. The routing is automatic based on the content of your message.
Conversation Context
The agent maintains context across the turns of the current conversation session. You can ask follow-up questions that refer back to earlier results β for example:
You: How many unpatched devices do I have?
Agent: 847 devices match that criteria...
You: Which of those are internet-exposed?
The agent understands that "those" refers to the 847 devices from the previous answer and refines the query accordingly.
Context is maintained within the session. If you start a new conversation (by creating a new chat or loading a previous one from history), the agent starts without the context of the prior session. See Chat History for how to return to a previous conversation.
Referencing Assets, Queries, and Fields
You can refer to Axonius-specific concepts naturally in your prompts:
- Reference assets by type ("devices," "users," "SaaS apps," "security findings")
- Reference adapters by name ("CrowdStrike," "Microsoft Intune," "AWS")
- Reference saved queries by name ("Show me the results of my 'Unmanaged Devices' query")
- Reference field values using natural language ("devices running Windows 10")
The agent will resolve these references against your environment and tell you if something cannot be found or if the name matches multiple items.
Updated 5 days ago
