How it works

Agents propose. Humans dispose. The executor materialises.

That single constraint is the whole design. It is enforced at the database, not just in application code: an aib_agent role can only insert proposals, a separate aib_app role performs the writes. An agent that tries to do anything else gets a permission error — the structural answer to the governance question every regulator is asking about automated decision-making.

What it covers

Client lifecycle

Onboarding, KYC, UBO/PEP, sanctions, MiFID suitability, GDPR data rights.

Banking

Deposits, lending (Lombard + IFRS 9 ECL), payments (SWIFT/SEPA-Instant), cards, FX.

Markets

Trading, settlement, custody, corporate actions, securities lending, derivatives.

Wealth & funds

Discretionary mandates, asset management (AIFMD/UCITS), pensions, fiduciary structures, insurance wrappers, crypto/CBDC custody.

The general ledger

Double-entry, balanced, append-only, period-closed, in integer minor units.

Regulatory reporting

The EBA DPM metamodel projected live from the GL into COREP, FINREP, LCR, NSFR, AnaCredit, EMIR, SFTR — validated and written as xBRL-CSV.

Controls

Reconciliation, break aging, governed period close, DORA incident handling, a daily control calendar.

Agent governance

A live model-risk gate that suspends an agent whose override / SLA / hallucination metrics breach, resumed only on human sign-off.

AIB is honest about its edges. The double-entry ledger ties out, the prudential reports compute live from the GL, and the execution plane, audit trail and model-risk gate are all enforced and tested. Connectors are mock by default — there is no live SWIFT/SEPA/SIC, no CSD/custodian, and no live filing channel to a supervisory authority. See what’s real and what’s simulated.