OS // Architecture

AI OS reference architecture for a trading brokerage.

The abstract stack, made concrete. What every layer of an AI-First OS actually looks like inside a live brokerage.

Every brokerage we work with wants the reference architecture — the concrete answer to 'what does this actually look like at our shape of firm?'. Below is the pattern we ship. Every element is deliberate, and every element maps to one of the six OS layers.

Data (L0) — what we connect

  • CRM (leads, clients, KYC, complaints)
  • Trading platform (positions, PnL, volumes, deposits, withdrawals)
  • Payment provider (deposits, chargebacks, funding source)
  • Comms (email, chat, call transcripts)
  • Marketing (paid channels, attribution, retention campaigns)

Model Router (L1) — the shape

  • Private-hosted open-weight tier for anything touching client PII or trade data.
  • Frontier API tier for public-web research and general reasoning.
  • Small local classifier for policy decisions and routing.

Firm Memory (L2) — the scopes

  • Client scope: KYC file, deposit history, complaints, trades.
  • Compliance scope: house rules, jurisdictional matrices, latest regulator communications.
  • Marketing scope: campaigns, creatives, compliance sign-offs.
  • Ops scope: runbooks, incident history, vendor SLAs.

Workflows (L3) — the first four

  • Onboarding & KYC — every new client, every jurisdiction.
  • Deposit funding anomaly — every deposit above a threshold or outside a pattern.
  • Retention / dormancy — every client hitting an inactivity trigger.
  • Complaints triage — every inbound complaint, classified, routed, drafted.

Agents (L4) and Skills (L5)

  • One agent per workflow, scoped to that workflow's tools.
  • Skills stack per desk: onboarding (KYC review, sanctions cross-check, funding-source classification), retention (dormancy triage, win-back drafting, next-best-action), compliance (marketing pre-check, complaints triage, regulatory-change monitoring).

Where to start

Every KJ Capital engagement begins with a two-week £15k AI Diagnostic. We map your existing data estate, model your firm's OS across the six layers above, cost the build, name the first three skills to ship, and hand you a plan your board can act on. It is deliberately priced to be a rounding error against the value it unlocks.

FAQ

How long to ship this?

Six to twelve weeks for the first shippable slice — one workflow (usually onboarding), two to three skills, memory and router underneath. The four-workflow shape is typically six to nine months.

Does the trading platform have to be replaced?

No. The OS connects to the trading platform. It does not replace it.

Start with the £15k AI Diagnostic

Two weeks, one costed OS blueprint, three named skills to ship first.