Asking Questions About Your Assets

Type your question in the input field the same way you would ask a colleague. You do not need to know AQL, field names, or which adapter the data came from.

A few examples of the kinds of questions you can ask:

  • "How many Windows devices haven't been seen by any agent in the last 30 days?"
  • "List all users with admin privileges who haven't logged in this month."
  • "Which devices are running an end-of-life operating system?"
  • "Show me all SaaS applications with a high-risk rating."
  • "How many critical security findings are currently overdue?"

The agent interprets your question, determines the appropriate filters and asset type, and returns the results. For questions that return asset records, you will see a count and a table with links that open the corresponding records in Axonius.

Supported Asset Types

The AI Companion can query all major asset types in Axonius, including:

  • Devices — laptops, servers, virtual machines, network devices, cloud instances, and more
  • Users — directory accounts, SaaS user accounts, and identity records
  • Software — installed applications and packages on managed devices
  • SaaS Applications — applications discovered through SaaS adapters
  • Security Findings — CVE matches, vulnerability scan results, and other security findings
  • Vulnerabilities — aggregated CVE and exposure data

The agent also supports queries across other asset types configured in your environment. If you ask about an asset type and the agent cannot answer, it will tell you why — for example, if the relevant adapter is not connected.

How the Agent Translates Your Question

When you ask an asset question, the agent uses a natural-language-to-AQL tool to translate your intent into an Axonius Query Language (AQL) filter. It selects the correct asset type, identifies the relevant fields, and applies any conditions you described.

If your question is ambiguous — for example, if the same name matches multiple fields or if a specific value does not appear in the data — the agent will ask a clarifying question before running the query. Once the question is answered, it proceeds with the refined query.

The AQL filter itself is not shown in the chat. The agent presents the results in plain language and table form.

Refining a Result

After the agent returns a result, you can ask follow-up questions in the same conversation to refine or explore further:

  • "Of those, which are exposed to the internet?"
  • "Break that down by operating system."
  • "Show me just the ones with a severity of Critical."
  • "Which adapter reported those devices?"

The agent maintains the context of the current conversation, so it knows what "those" refers to from your previous question. You can keep refining through multiple turns until you have the answer you need.

If you want to save the underlying query the agent built, ask it: "Save this as a saved query called [name]." The agent will create the saved query in Axonius and confirm when it is done.

Query Cards and Asset Table Widgets

When the agent returns a result based on a saved query, it shows a query card with an Open in Assets Page link. Select that link to open the full asset list in Axonius with the query pre-applied.

Some results appear as an asset table widget — for example, "Top 10 Active Managers by Start Date" — rather than as a query card. Asset table widgets don't include an Open in Assets Page link. To explore the full results, ask the agent for a saved query or navigate to the asset page directly in Axonius.

Checking Whether Devices in a Coverage Gap Are Really Missing

When you ask which devices are missing an agent, for example "Which Windows servers are missing CrowdStrike?", a device can appear in the gap because the same machine reached Axonius as two records that didn't merge. The record without the agent shows as missing, although the machine is covered.

Axonius AI checks the gap for these duplicates when you ask about a missing agent for a specific adapter, or when you ask it to find under-correlated devices in the gap. It compares each device in the gap with the covered devices, and splits the gap into three parts, each with a count and a percentage of the gap:

  • Likely the same device as a covered one — A strong or medium match with a covered device.
  • Weak match — A match on an IP address only, or on a value that devices share.
  • No match — Devices that are really missing the agent.

Consider the following:

  • Axonius AI matches devices on hostname, asset name, MAC address, serial number, and IP address. It ignores differences in letter case and domain suffix when it compares names.
  • Axonius AI shows the original gap first, and then how many devices in it are likely the same device. It doesn't subtract those devices from the coverage number, and it doesn't count a weak match as coverage.
  • Axonius AI names two or three sample pairs, with links to the device profiles and the values that each pair shares. Ask Why Didn't These Merge? to see why a pair didn't link.
  • When you ask for the check directly, Axonius AI can tag the likely duplicate devices or save them as a query. Axonius AI asks for your confirmation first.
  • The check covers devices only. It reads up to 25,000 devices in the gap and up to 150,000 covered devices. When the check reads only part of the gap, the answer says how many devices it checked.
  • When no covered device exists to compare with, the check doesn't split the gap. The answer says that it couldn't split the gap.

The scheduled Coverage check worker in the AI Workforce reports the same split in its briefing, for example "Likely the same device".

Checking Whether a CVE Affects You

Ask whether a specific CVE affects your environment, for example "Am I affected by CVE-2026-12345?". The AI Companion reads the advisory first, and then shows two separate lists, each with its own count:

  • Scanner findings — The devices on which a vulnerability scanner reports the CVE.
  • Installed software — The devices that have a vulnerable version of the affected product installed, based on the installed software data.

Consider the following:

  • When both lists are empty, the answer is "Not affected on current data". The answer names the CVE IDs, the fields the AI Companion checked, the product names it looked up, and the date of the data. A scanner can report a new CVE late, so check again later.
  • When the scanner list is empty and the installed software list isn't, the AI Companion doesn't answer "not affected". It uses the installed software list as the affected devices.
  • When no vulnerability scanner is connected, the scanner list can miss the CVE. The installed software list is the main evidence.
  • For a product that no connected source can see, for example a network appliance, firmware, or a self-managed server application, the answer is "not visible to Axonius".
  • When no device is affected, the AI Companion doesn't build a chart or dashboard of zeros. It offers to create a private saved query or dashboard that fills when a scanner reports the CVE.
  • The AI Companion uses only internet-facing data to judge internet exposure, for example an external scan, Internet Exposure from Axonius Exposures, or a verdict from a connected cloud adapter. A public IP address isn't evidence of exposure. When none of these fields exist, the AI Companion says that internet exposure isn't available from your data.

Did this page help you?