Skip to content
The Platform

One platform to see, govern, and act on AI usage.

From detecting risk in a prompt to applying policy, supporting human review, and surfacing patterns over time — Veytrio brings AI governance together in one place.

Request a pilotSee how it works
  • Risk detection
  • Policy engine
  • Human review
  • Behavioural intelligence
Risk AnalysisLive

Prompt captured

Push config with key sk-live-8842 for customer Northwind account AC-40192.

Detected

Credentials & secretsCustomer data

Severity

InfoLowMedHighCrit
Policy · Credentials ≥ HighEscalate to admin
Detection & risk intelligence

The anatomy of a flagged prompt.

Veytrio inspects prompt activity for the content that creates governance risk, categorises it, scores it by severity, and explains why — so reviewers can focus on what matters. Follow one event from capture to escalation.

  1. 01

    Captured & redacted

    Sensitive fields — names, emails, account numbers, credentials — are redacted before storage or review.

  2. 02

    Categorised & scored

    Every event is rated across five severity levels — Informational, Low, Medium, High, or Critical — and sorted into risk categories.

  3. 03

    Matched & explained

    Each event carries a plain-language reason for its rating, so the decision is transparent — then policy decides what happens next.

Prompt capturedLive

“Draft a renewal note to Helen Mwangi at Northwind Trading, account AC‑40192 — and paste the API key sk‑live‑8842.”

4 sensitive fields redacted before storage
Categorised & scoredLive

Detection categories

  • Sensitive personal data
  • Credentials & secrets
  • Source code & IP
  • Financial information
  • Healthcare & special-category
  • Legal & regulatory
  • Customer & client confidential

Severity — five levels

Info
Low
Med
High
Crit
Scored High · 2 categories matched
Matched & explainedLive
Credentials & secretsatHigh→Escalate to admin

Why it was flagged
A live-looking credential appeared in an outbound prompt alongside customer account details — above the policy threshold for this category.

Escalated to admin · evidence record kept

Illustrative example · synthetic data

Detection uses high-quality pattern and keyword matching — a strong signal to prioritise review, not a guarantee that every risk is caught.

Policy engine & review

Turn risk signals into governed action.

Veytrio gives you an organisation policy engine and human review workflows — so the right activity is flagged, routed, and resolved by the right people.

Six controls · one rule composer

Rule composerIllustrative
ConditionIF
Credentials & secrets

Risk category

Threshold≥
High

Minimum severity

ActionDO
Escalate to admin

Policy action

If Credentials & secrets reaches High then Escalate to admin.

Named AI policies

Define policies by risk category, minimum severity, and the action to take — from logging to escalation.

LogAlertEscalateAcknowledge

Sensitivity tuning

Enable, disable, or tune each detection category (lenient, standard, or strict) to fit your risk appetite.

Policy library

Start from ready-made policy templates and apply them in a click, then adapt to your organisation.

Reviewer decisions

Authorised reviewers mark events reviewed, dismissed, or escalated — with a full audit trail.

False-positive workflow

Flag false positives and give structured feedback — the correct category or severity — to sharpen review.

Escalation & acknowledgement

Escalate critical activity and capture policy acknowledgements, so accountability is recorded.

Policy actions support your team — they flag, alert, escalate, and request acknowledgement. People decide what happens next; Veytrio keeps the record.

Behavioural intelligence

Patterns surfaced for people, not decided by machines.

Veytrio looks across activity over time to highlight where risk concentrates, always keeping a human in the loop.

  • Repeat-risk patternsIdentify repeat high-risk activity, repeat policy violations, and category trends across teams.
  • Anomaly detectionSurface critical spikes, new-tool adoption, and unusual sharing of source code or credentials for human review.
  • Prioritised review queueRank what to look at first by severity, sensitivity, repetition, and recency — with an explanation for each.
  • Advisory recommendationsReceive suggested policy or escalation actions an administrator can accept, modify, or reject — never applied automatically.

Human-supervised by design. Behavioural insight is advisory only, supports a privacy-minded anonymised executive view, and never produces automated decisions about individuals.

Prioritised review queueRanked
  • 01
    HighCredentials & secrets

    Repeat high-risk activity — 3 events this week

  • 02
    HighCustomer & client confidential

    Spike above this team’s baseline

  • 03
    MediumSource code & IP

    Unusual sharing to a newly-adopted AI tool

  • 04
    LowFinancial information

    Single flagged prompt — first occurrence

Illustrative example · synthetic data · human review required

Coverage

Where Veytrio captures AI activity.

Veytrio runs on one source-agnostic event model, so every channel flows into the same review, policy, and reporting pipeline.

Three channels · one pipeline

01

Browser tools

Available

A managed browser extension captures activity from ChatGPT, Claude, Gemini, and Microsoft Copilot.

02

VS Code (IDE)

Early

An early VS Code integration captures explicit developer prompts and selections through the unified event model.

03

Internal AI tools

Push connector

A source-agnostic push connector lets internal chatbots and LLM gateways send governance events to Veytrio.

Capture is primarily browser-based today; IDE and internal-tool coverage is expanding. Coverage depends on your deployment and supported environments, and Veytrio is designed for governance visibility over managed usage — not to be presented as impossible to bypass.

See Veytrio against your own AI usage.

Run a controlled, synthetic-first pilot and see detection, policy, and review working end to end.

Request a pilotBack to overview