Use case · Find Every AI
Finish the AI inventory your other tools started.
Every AI app, model and local runtime in use across your environment — including the AI nobody declared — discovered from the security tools you already run and ranked by which critical systems it touches, with the person behind it named where your identity data supports it. Exports as evidence an auditor will accept. No agent, no proxy, nothing installed.
Every approach sees one slice.
The AI you're missing isn't hiding. It's in the slice your tool was never built to watch. Browser tools are excellent inside the browser. Gateways are excellent on traffic that crosses them. Neither was built to look at what is installed on the machine.
| Where AI actually shows up | Browser extension | Network gateway | SaaS posture tools | KeyCaliber |
|---|---|---|---|---|
| Web AI apps used in a managed browser | ● | ● | ● | ● |
| AI called from servers, scripts and APIs | — | ● | — | ● |
| AI features inside apps you already approved | — | — | partial | ● |
| AI models running locally on a laptop | — | — | — | ● |
| Which of that AI touches critical systems | — | — | — | ● |
| Who approved it, when, and what has changed since | — | — | partial | ● |
Seven ways KeyCaliber finds AI.
No single feed is mandatory, and coverage scales with what you connect. Found five ways, an app is still one row.
Network and firewall logs
Traffic to AI services, scored for confidence.
DNS logs
Resolves destinations that raw traffic data leaves unidentified.
Web gateway and CASB
Full request detail, including the model in use.
Single sign-on events
Names the person behind the login, where your identity data supports it.
Installed software
Local AI runtimes on laptops and servers, read from the EDR and MDM inventory you already collect. Every vendor's, not just one — and tied to the same asset as everything else we find.
Embedded-AI catalog
Which approved vendors ship AI, whether it's on by default, and whether they train on your data.
Manual entry
Add what your team already knows. Discovery merges into it rather than around it.
A list becomes a work queue.
Forty AI apps is a research project. The same forty, ordered by what they touch, is a short list your team can finish this quarter.
- Act now
Local AI runtime on a finance workstation
1 person · asset identified as critical · prohibited software
- Act now
Consumer AI app used from a payroll server
3 people · vendor trains on submitted data by default
- Review
AI writing features switched on in an approved app
Whole tenant · enabled by default at the vendor's last release
- No action
Approved AI assistant, engineering group
61 people · inside the approved population
Illustrative findings
Then the list has to hold up. Every row carries a decision, an owner and the history behind both, and exports as the register an auditor asks for.
Govern the AI you find →Find the AI you don't know about.
Connect one endpoint tool and one network source. Most first runs surface local AI runtimes and approved-vendor AI features nobody knew were there.
Request a demo