The system of record for your agency's AI use.

Your staff already use AI, and the agency keeps no copy of what they send. Verillian is software on their computers that applies your AI policy and keeps a record of each request it captures on your own servers.

Try it: pick a request and see your policy check it on the device.

Each request it captures leaves a record on your own servers.

Draft a reply to the resident at 555-0142 about her permit.

Redacted

What the provider gotDraft a reply to the resident at [PHONE] about her permit.

The phone number never reached the AI provider.

Signed on the device. Kept on your own servers: 10:43, a records clerk sent ChatGPT a reply to draft, with the phone number replaced. Redacted.

Do your records show how staff use AI?

Staff paste emails, notes and reports into AI tools to save time. Then someone asks what was sent, by whom, and whether it was allowed. Without a record, the agency has nothing to show.

82%of state CIO organizations say their staff use generative AI in daily work, up from 53 percent a year earlier.NASCIO, The 2025 State CIO Survey

How does Verillian work?

Person and AI appYour staff
Verillian softwareOn each device
  • Allowed
  • Redacted
  • Denied
AI providerOutside your agency
The recordOn your own servers
  1. Install

    Your IT team installs Verillian on staff computers and runs the console on a server you choose.

  2. Check

    Before a question is sent, Verillian checks it against your policy. On the services it inspects, flagged details become placeholders.

  3. Record

    Each request it captures is recorded on your own servers, encrypted to your organization's key.

Watch one request get checked and recorded, live.

What changes when an AI tool acts on its own?

Some AI tools now send email, run commands and delete files on their own. For tools like Claude Code, your policy decides which of those actions run.

Without Verillian

Sent

  1. What the person asked
  2. >Draft a reply to this resident: "My refund is three weeks late. You can call me at 555-0142 if you need anything."
  3. What the AI did
  4. send_email(subject: "Your refund")
  5. Sent
  6. Done. I sent the resident a reply.
With Verillian

Denied

  1. What the person asked
  2. >Draft a reply to this resident: "My refund is three weeks late. You can call me at 555-0142 if you need anything."
  3. The provider receives: [PHONE_REDACTED]
  4. What the AI did
  5. I attempted to call the `send_email` tool with input {"subject":"Your refund"}. This tool call was blocked by policy and was not executed.
  6. I could not send it. The reply is drafted for you to review.

Is it built for organizations like ours?

Every county, city and state agency answers to its own rules. Pick yours to see how Verillian fits the way you work.

What do IT teams ask first?

How do we set it up?

Your IT team installs Verillian on staff computers and runs the console on a server you choose: on premises or in a private cloud you run. Runs on macOS today, with Windows and Linux in early access.

Which AI tools does it work with?

For example, ChatGPT and Claude Code. Verillian works at the device, so there is nothing to set up inside each AI tool. On the services Verillian inspects, a request it cannot read is refused, and each request it captures is recorded.

What happens when a request breaks policy?

Verillian checks each request on the device before it is sent, and for tools like Claude Code, each tool call before the tool runs. When a rule redacts, the AI provider receives only a placeholder.

Where are records kept, and who can read them?

Records stay on your own servers, encrypted to your organization's key, which only your agency holds. Reviewers you authorize open records in your console, and each open is recorded. Verillian never receives that key.

Ready to see it on a real device?

Tell us about your agency on a short form, and we set up a demo. We open the signed record together. Verillian never receives your prompts, responses or records.