Sarghaby Araghatta
Request a free assessment →
Home / Sensor

Find every AI tool on your endpoints, without slowing them down.

One light agent for Windows, macOS and Linux finds the AI in use and masks secrets on the device before anything leaves. It observes by default and blocks only when you switch that on.

Your environment
Endpointlaptop or server
Browser extensionChecks text you submit to an AI tool.
Optional
Sargha sensorFinds AI, read-only. Masks secrets on the device.
  • Agents
  • Tool servers
  • Keys
Device managementPushes the sensor out.
Sargha cloud
CollectorReceives over an encrypted connection.
  • Authenticates each machine
  • Keeps your tenant's data apart
ScoringTies each find to an owner and rates risk.
Console and reportsWhere the work happens.
  • AI inventory and coverage
  • Ranked fixes with owner and reach
  • Signed evidence and reports
Your tools
SIEMSplunk, QRadar, Sentinel, Elastic, Chronicle
TicketsJira, ServiceNow
ChatSlack, Microsoft Teams
  • Metadata
  • Findings and alerts
  • Identity and access
  • Evidence and reports
  • Dashed: optional
  • Thicker: more data
  • Encrypted in transit
  • Read-only
  1. 1
    Install and enrollYour device management installs the sensor. The machine then holds a credential of its own.
  2. 2
    Inventory leaves the machineAI tools, agents, tool servers and masked keys, with machine, user and account.Never sent Key values, file contents, prompts.
  3. 3
    Checked textOnly the text being submitted, checked in memory and not stored.
  4. 4
    Attribute and scoreEach find is tied to its owner and rated for risk and live threat.
  5. 5
    Findings reach the consoleA ranked list of fixes with reach and owner.
  6. 6
    To your SIEMFindings with full context.
  7. 7
    To ticketsOne ticket per risk, assigned to its owner.
  8. 8
    To chatAlerts in the channel you choose.
Discovery is read-only. Secrets are masked on the device.

See what your EDR does not inventory

Coding agents, tool servers, keys, extensions and local models, matched against whether each machine runs endpoint protection. One view answers: which machines run AI with no endpoint protection?

Light enough to leave running

Measured on one Mac and one Linux VM, it uses a tiny share of one core and slows itself down if it ever uses more. Windows, bare-metal Linux and long runs have not been measured yet.

Private by design

Read-only by default. It never reads prompts, chats or file contents, and key values never leave the device: only a masked start and the length.

Observe first, enforce only if you choose

Warn, then block or quarantine, per policy, with every action authorized by an operator and recorded. Servers, production and unknown machines are never isolated automatically.

Rolls out with the tools you have

One self-contained program per operating system, pushed through Intune, Jamf, Kandji, SCCM or Ansible, and removed with one command.

Governs AI-written code

Detects AI coding tools and the repositories they wrote, then keeps a review state, takes in the results of your own scanners and records the engineer's sign-off.

What runs where, stated plainly

  • AvailableDiscovery, scoring and evidence on Windows, Windows Server, macOS and Linux.
  • Linux onlyLive tracing of programs and connections is Linux only today. macOS and Windows collect on a schedule. Roadmap
  • AvailableBlock and quarantine on Windows, macOS and Linux. They switch on only with the Enforcement licence and your opt-in, are checked on the machine after they apply, and are fully undone when you switch them off.
  • AvailableChecks of what a person submits to ChatGPT, Claude or Gemini, through the managed browser extension on user workstations in Chrome and Edge (checked in memory, not stored), and server-side checks through an inspection service you run.
  • AvailableLinux builds (GPG), Windows builds (Authenticode, EV certificate) and macOS builds (Apple Developer ID, notarized) are signed today.
How it works

Step by step.

  1. Deploy

    Push one self-contained program per operating system through your device-management tool.

  2. Observe

    It reads the machine on a schedule, finds AI, masks secrets and reports. Nothing is blocked.

  3. Choose to enforce

    Where you decide to, switch on warn, then block or quarantine, per policy.

Questions

Common questions.

What does the sensor send?
Which AI tools, agents and tool servers are present, whether a key exists (a masked start and its length), and device details. The sensor never sends key values, file contents, prompts or keystrokes. The optional browser extension checks submitted text in memory and does not store it.
Does it need administrator rights?
The service runs with elevated rights because discovery must read other users' files. Enforcement actions also need them, and every one is authorized by an operator and recorded.
What you need
  • A way to push software to your machines.
  • Administrator or root rights for the service, because discovery has to read other users' files.
  • An outbound secure connection to your Sargha collector.
For your technical team

Schedule, network destinations, signing chain, privileges per operating system, footprint and the platform matrix are in the technical pack, shared with your team under NDA.

Request the demo and technical pack
Get started

See the AI on your fleet in about a week.

Run the sensor across your estate and receive an AI risk report and an AI Bill of Materials. The assessment is free and carries no commitment.

Request a free assessment → Book a 30-minute call