AI‑native operations for private wealth

AI plans.
Rules decide.

SERAI is the operations platform for private banks, boutique investment houses and family offices. It builds the screen each person needs, the moment they need it — and never lets a model approve anything.

The problem

Wealth managers run on software built for someone else

Every screen is a project

A new report or field means a change request and weeks of waiting.

Every integration is a negotiation

Custodians, data vendors and regulators each arrive with their own format.

The market moves faster than the desk

A rate shock lands on Monday; who is exposed is known on Thursday.

AI plans. Deterministic rules decide.

The AI proposes

screens, plans, connectors and advice

The policy kernel decides

same input, same answer, every time

  • PASS goes to a human approver
  • FAIL stopped, with the rule cited
  • UNKNOWN also stopped: missing data never passes

Two cases

The screen is built for the person and the situation in front of them

Case 1 · Customer onboarding

A family trust arrives

  1. The request is read and the client is classified as a discretionary trust across four jurisdictions.
  2. Every party is screened against sanctions and politically exposed person lists. One beneficiary comes back UNKNOWN.
  3. UNKNOWN blocks. The screen is composed for that exact situation: trust structure, source of wealth, a compliance review panel, a rule‑set risk rating and maker‑checker approval.

A clean individual would get a two‑panel screen instead.

Case 2 · Trading desk

“Hedge most of our EUR, keep some upside”

  1. Net EUR exposure is read from the book, every figure cited to its source positions.
  2. The AI proposes four hedges. The mandate rules strike out the one that is too large and the instrument that is not approved.
  3. The ticket is composed and approved by a supervisor with a single‑use token. The AI tries to approve it and is refused.

A conservative USD‑only mandate would get one line: “No EUR exposure to hedge.”

Try it yourself

Type a request, switch between a relationship manager, compliance officer, portfolio manager and head of risk, and watch the same request produce different screens.

Open SERAI Studio

How it works

One truth at the centre; everything else orbits it

Nothing in the orbit can write without passing through the core. The same runtime is designed to deploy as SaaS, in a bank’s private cloud, or fully air‑gapped; packaging for all three is on the roadmap.

World model
One event‑sourced record of clients, mandates, positions, cash and obligations. History is never rewritten.
Policy kernel
PASS, FAIL or UNKNOWN. Maker‑checker gates and approval tokens bound to the exact proposal. Never consults a model.
Evidence chain
Every figure cites its source; every decision replays identically.
World modelPolicy kernelEvidence chain PlannerRequestsComposerForgeSentinelLearning

Progress · as of 30 October 2026

The spine is built and tested, not described

  1. World modelMerged · 3 reviews
  2. HardeningMerged
  3. Policy kernelMerged
  4. CapabilitiesMerged · 33‑probe attack battery
  5. Market sentinelIn review · 10/10 tests
  6. End‑to‑end demoReady · tests written first
78automated tests passing
0runtime dependencies
3 daysfrom empty repository to four merged components
M1 · nowWalking skeleton: fail‑closed control and market sensing
M2Faces: screens built on the fly
M3Edges: integration on the fly, first pilot
M4Growth: self‑improvement under governance

Where it comes from

Born in Hong Kong and Shenzhen

Hong Kong brings the international mindset, the wealth hub and the clients. Shenzhen brings implementation capability and engineering speed. SERAI is cross‑border by design.

ShenzhenHong Kong

We are raising a seed round

We are looking for strategic investors with exposure to private wealth, and for banks and family offices who want to shape the first pilots.