L1
Sandbox
A capability starts here. It runs in a sandbox with synthetic, non-sensitive data only. It touches no real customer data and no production system. Most of the portfolio sits here, and we say so.
Fidevo Group · Governance
Governance: we state claims at the level of evidence we hold.
Fidevo builds financial-adjacent infrastructure and operational software. In that setting, the difference between what exists and what is planned matters more than the pitch. So we run an internal discipline, the AI Delivery Loop, that decides what a capability is allowed to do, and we publish the result of it as a plain status on every business line.
Trust levels
Every capability begins at Trust Level 1. Promotion is a decision backed by recorded evidence, with an owner, a source and a timestamp, not a judgement call made in the moment.
L2
Supervised with real data
Promotion happens only when there is a recorded evidence trail: what was tested, by whom, against what, and when. At this level a capability may see real data, but a person reviews its output before it has effect.
L3
Reduced supervision
Reduced supervision is earned, never assumed. It requires sustained evidence at the level below, a named owner, and a defined way to reverse the decision if the evidence stops holding.
ENGRAM
ENGRAM is the framework that decides what may become an AI skill or agent at Fidevo, and at what trust level. Before anything is built, ENGRAM asks what the capability is for, what data it would need, what happens if it is wrong, and who owns the consequence.
A capability that cannot answer those questions does not get built. One that can is admitted at Trust Level 1 with its boundaries written down, so that any later promotion is measured against something specific rather than against a memory of the original intent.
ENGRAM is in daily internal use today.
AI Guardian
AI Guardian reviews Fidevo's own AI-assisted work. Where ENGRAM governs what may be built, AI Guardian looks at what was actually produced: whether a claim is supported, whether a status has been quietly upgraded, whether an output reflects the evidence behind it.
That review covers this website. The statuses on the business line pages exist because of it, including the ones that say a product is specified but not built.
AI Guardian is in daily internal use today.
What this means when you read this site
A business line marked Live has confirmed real usage today. Where only part of it is live, the page says which part.
Early access means a real technical build exists and access is granted on request. It does not mean generally available, and it does not imply pricing.
Design intent means the capability is named and specified but not yet built or proven live. Where a name is provisional, we say that too.
We do not publish customer counts, certifications or capability claims we cannot evidence. If a figure is not on this site, it is because we do not currently hold it.