Always-on risk
The problem you would have found in three weeks, today.
Sentinel is an always-on risk radar that sweeps your deals, documents, counterparties, provenance and book in the background, and tells you what it found in plain language. Thirty-six checks, nine families, and detection that is pure deterministic SQL rather than a model guessing. It never blocks an action: Sentinel flags, you decide.
36
Background checks
9
Risk families
Deterministic
Detection, not guesswork
Advisory
Never blocks a trade
Why it matters
Built to be believed.
Risk that only shows up when things are joined
A counterparty record looks fine. A deal looks fine. That counterparty on that deal is the finding. Cross-entity checks are the ones a spreadsheet structurally cannot make.
A state, not a stream of alerts
A finding is one live row that resolves itself when the condition clears. It does not nag, it does not stack duplicates, and dismissing one is permanent.
Deterministic first, explained second
Every detection is SQL over your own data, so it costs nothing and cannot hallucinate. AI is used to explain a handful of findings and to draft a fix, never to decide one.
Coverage
Thirty-six checks across nine families
Compliance, cross-entity, deals, documents, provenance, trust, portfolio, account limits, and inbound messages. The full catalogue is published, because a risk tool that will not say what it checks is asking for faith.
- Sanctions hits left unresolved, screenings that have aged past 90 days, and lists that stopped updating
- Deals that contradict themselves across their own documents, parties and passports
- Certificates expiring, documents filed as the wrong type, evidence photos with edit signals or the wrong country in their GPS
- Passports that are unsealed, have no custody chain, or are missing required evidence
Soundness analysis
Low risk78
Completeness · consistency · counterparties
Per deal
A compliance officer on every deal
Four dimensions scored from real data: sanctions and AML, KYC and KYB, due diligence against the commodity's OECD pack, and chain of custody. The verdict is rule-based and auditable, so it can be defended rather than merely trusted.
- Clear, review, or blocked, decided by rules and never by a model
- Grounded in the actual participants, passports, checklists and custody events on that deal
- An AI rationale explains the picture, and is cached until the picture moves
- Bridged into the radar, so a deal that stops being clear raises a finding
Screening result
Confidence 0.86
2 matched entries · history saved
Response
It flags, you decide, and it remembers
Acknowledge, snooze to a date, or dismiss permanently. Nothing Sentinel finds ever blocks an action, and nothing it drafts is ever sent on your behalf.
- Findings auto-resolve when the underlying condition clears
- Draft-a-fix writes a ready-to-use reply or note for you to review and send yourself
- Severity decides reach: informational stays in-app, warnings and critical findings also email
- Delivery mirrors permissions, so a finding only reaches people already entitled to see the thing it is about
AI assistant
Memory + voiceCIF includes freight and insurance to the named port. CIP extends that to any transport mode.
How it works
From start to proof.
Turn it on
An owner or admin enables Sentinel for the organisation in settings. It is included on every paid plan.
It sweeps in the background
Scans run nightly and after meaningful changes, debounced so nothing runs on top of itself.
Findings arrive with their reasoning
Each one states the deterministic detail, links to the entity it concerns, and suggests a next step.
You decide
Act on it, acknowledge it, snooze it, or dismiss it for good. Sentinel never takes the action for you.
FAQ
Common questions.
Does Sentinel ever block a deal or a document?
No, and that is a design decision rather than a limitation. It is advisory throughout: it raises findings, explains them, and suggests next steps. Every action stays available to you. A tool that blocks trades on advisory signals gets switched off within a month.
Is this AI deciding things about my business?
No. Every detection is deterministic SQL over your own data. AI is used in exactly two places: to write a plain-language explanation on a handful of finding types, and to draft a fix you asked for. The deal compliance verdict is rule-based and auditable. If AI is unavailable, detection carries on unchanged and the deterministic detail still shows.
What does it actually check?
Thirty-six checks across nine families, all published in the signal catalogue. Detection is free and always on; a credit is spent only when Sentinel writes an explanation or drafts a fix.
Who sees a finding?
Delivery mirrors the permissions on the underlying record. A finding about a deal reaches the people on that deal; one about a member's own account reaches only them. Sentinel cannot surface something to someone who could not already see it.
What happens when I fix the problem?
The finding resolves itself on the next sweep. A finding is a statement of a current condition, not an event log, so a resolved condition means a resolved finding without you closing anything.