The roles
A viewer sees dashboards and coverage. Billing adds invoices. An analyst can open individual events. An admin also sets policy, roles and the fleet controls.
Overview
The first screen an admin signs in to
Five counts across the top, then every finding broken down by framework, by risk and by what the device did about it, then the most recent events. The window runs from the last 24 hours to the last 90 days.
A rise in events is drawn in cherry, because more sensitive data reaching AI tools is the number the product exists to bring down.
Overview
1,842 devices reporting · last event 4 minutes ago
Events
2,417
↑ 12% vs previous 30 days
Blocked before send
311
12.9% of events
Redacted and sent
286
11.8% of events
Warned
538
22.3% of events
Devices reporting
1,842
99.0% of fleet
Findings by framework
Risk mix
What was done
Recent events
| When | User | Risk | What was found | Policy | Client to destination | Action |
|---|---|---|---|---|---|---|
| 14:22 | r.okafor | Critical | Card number | PCI DSS | From macOS, Chrome to ChatGPT in a browser tab. | Blocked |
| 14:09 | m.delacroix | High | Person name in context | HIPAA | From macOS, Safari to Claude in a browser tab. | Redacted |
| 13:58 | j.whitfield | High | Credential | Internal | From Windows to ChatGPT through its desktop app. | Blocked |
| 13:51 | a.nakamura | Medium | Email address | GDPR | From Windows, Chrome to Copilot in a browser tab. | Warned |
| 13:30 | t.abadi | Low | Your own rule: Deal codename | Internal | From macOS, Chrome to Claude in a browser tab. | Logged |
Illustration. The people and figures are invented. There is no column for the message because the console never receives one.
One event, opened
What an analyst sees when they open an event
The person, the device and browser it left from, where it was going and whether that tool is one you approve. Then each finding, with its framework, its confidence and a masked snippet, and the risk and the action the device took.
Risk and action are shown as separate facts. A Critical event that was only logged is not the same as one that was stopped, and a page that blurs the two lets someone believe the second when the first happened.
Only the analyst and admin roles can open an event, and every read is written to the audit log under the reader’s name.
Incidents
The alerts that need a decision, oldest first
A queue worked newest first starves its tail, and the alert nobody got to is usually the one that mattered. So the console shows how long the oldest has been waiting, not only how many there are.
Give each alert an owner, singly or in bulk, and filter to the ones nobody has taken or the ones assigned to you. An alert moves from open to investigating, and ends resolved, a false positive or a duplicate.
Incidents
14 waiting · the oldest for 3 days
| Waiting | Person | Risk | What was found | Policy | Owner | Status |
|---|---|---|---|---|---|---|
| 3 days | s.oyelaran | Critical | Bank account | GLBA | j.mensah | Open |
| 2 days | r.okafor | Critical | Card number | PCI DSS | Nobody yet | Open |
| 9 hours | p.rasmussen | High | National ID | GDPR | Nobody yet | Open |
| 40 minutes | j.whitfield | High | Credential | Internal | a.lindqvist | Open |
Illustration. The people and figures are invented.
Closing an alert asks why, from a fixed list
- Real, and it should not have been sent
- Real, but allowed
- Approved destination
- Not sensitive at all
- Wrong kind of identifier
- Test or seed data
- Already handled elsewhere
- Something else, with a note
The fixed list is what lets the console count the closes. The console totals the closes that blame detection over the last 90 days and splits them by finding, so “an employee ID read as a social security number, eleven times this month” shows up as a gap with a fix rather than as eleven annoyed analysts.
Policy
Set it for the organisation, then change it where a team needs to
Start from an industry template and adjust from there. A department changes only the settings it names, and everything it leaves alone keeps following the organisation, including later changes.
-
Each framework
Switch it on, choose whether a match is logged, warned, redacted or blocked, set the confidence at which it warns and at which it blocks, and decide whether tools you approve are exempt.
-
Departments
Sales or Engineering can enforce their own version. A save that makes a department more lenient than the organisation shows what it loosens, asks you to confirm, and is recorded under your name.
-
Your own rules
A pattern, or a kind of thing you name yourself, such as a project codename or a patient name, which the model on each device then looks for. Group your rules into a set of their own.
-
Industry templates
Healthcare, legal, finance, retail, education, professional services and technology. The preview shows what applying one would change, and applying a template only ever adds protection.
Also under Policies
- Redaction What a replaced value looks like
- Approved tools The AI tools you sanction
- Images, files and large pastes An action for each
- Sites left alone Banking and health sites where nothing is read
A framework here is a set of detectors and an action, not a certificate. Devices pick up a saved change on their next policy check, about once a minute, and every change is written to the audit log.
Coverage health
A device that stopped reporting looks exactly like a compliant one
On every other chart, zero findings and zero coverage are the same flat line. So a seat only counts as covered once its device has reported in, and the Coverage page lists what needs attention until the list is empty.
The detail splits into devices, sites and versions, down to which detector release each device is running.
What needs attention
1,842 of 1,860 devices reporting
- 23Seats with no device. People who never got the agent, and are invisible to every other number.
- 9Silent for over a day. No heartbeat, so nothing on that machine is being checked.
- 2Not binding on a site. The extension is installed, but the message box on that site was not found.
- 6Policy out of date. Still enforcing an older version than the one you last saved.
- 41Events refused. Sent, but rejected by the server, so nobody has them.
- 3Sent unchecked. A message reached an AI tool because the check took too long or no agent answered.
Illustration. The figures are invented. A revoked device is silent on purpose, so it is not listed here.
Access and audit
Who can open what, and a record of everyone who did
The rest of the console is about the people using it: what each role is allowed to see, how they sign in, and the log an auditor will ask for.
Single sign-on
Connect your own identity provider over OIDC, per organisation. Without it, sign-in takes a password and a code sent by email.
Audit log
Append-only. Policy changes, role changes, every read of an event and every staff access, exported as CSV or JSON for a SOC 2 or ISO 27001 assessor. The export is logged too.
Employee record
A person signed in to the console can see their own events, who opened them, and a list of what is never collected about them.
Fleet controls
Put every device, or one, into safe mode or pause capture for 30 minutes up to 24 hours. Each use is kept in a history.
Notifications
Alerts by email, Slack or Teams, each destination with its own minimum severity and verified with a code before anything is sent. A weekly summary of counts goes by email.
Shadow AI
AI websites people open that nothing watches yet, and AI tools installed on devices, MCP servers included. Each half has its own switch, and turning one on is logged with a reason.
Staff access
If our support staff need to see your console, they get a grant that expires, which you can be asked to approve, and it appears in your own audit log.
Upcoming
What is coming to the console, named before you ask
A security team will find these in the first week of a pilot, so they are listed here. Everything on the right is on the roadmap and none of it ships today.
In the console today
- Overview and events, with a page for each event
- Incidents, with owners and a fixed list of reasons
- Policy per framework and per department, with your own rules
- Coverage health for every device, site and version
- Notifications by email, Slack and Teams
- Audit log with CSV and JSON export
On the roadmap Not available
- SIEM export to Splunk, Microsoft Sentinel, a webhook or syslog with CEF
- Directory sync over SCIM, and SAML sign-in
- Tickets in ServiceNow or Jira raised from an alert
- Trend reports by team and by channel, and scheduled exports
- Evidence packs for HIPAA, GDPR, PCI DSS and SOC 2
- Time-limited exceptions with a named approver
Straight answers
What a security review asks about the console
Including the ones where the answer is not yet.
Can an administrator read what an employee typed?
No. The console shows the kind of data found, how many times, the confidence, a masked snippet, the destination, the risk and the action taken. The text itself was never sent, so there is nothing to reveal through a permission, an escalation or a support request.
Who can open an individual event?
People with the analyst or admin role. Viewers and billing users see dashboards and coverage but cannot open an event, and every read by an analyst or admin is written to the audit log under their name.
Does it send events to our SIEM?
Not yet. Export to Splunk, Microsoft Sentinel, a generic webhook and syslog with CEF is on the roadmap. Today, alerts reach your team by email, Slack or Teams, each with its own minimum severity, and a weekly summary goes by email.
How is one customer’s data separated from another’s?
Tenant identity is taken from the device’s mutual TLS certificate chain and checked on the server, never from a field the client supplies. Data is read under row-level security scoped to that one tenant.
Can we write our own detectors?
Yes. Add a pattern, or name a kind of thing, such as “patient name” or a project codename, and the model on each device looks for it. Your rules can be grouped into a set of their own and given an action like any framework.
How long are events kept?
A daily job deletes events once they pass your retention period, except those behind an alert that is still open or under investigation. The audit log is kept for seven years.