Inside the Agent
Pipeline

Copilot, Shield, Ghost, and Watch, in the order a transfer actually meets them, plus the API shape and latency budget behind them.

See Live Demo

One Call, Four Agents, In Order

Every transfer routes through the same pipeline: Copilot checks it against the user's baseline first. If it doesn't match, Shield scores it across six signal layers. If a user overrides a Shield block anyway, Ghost holds the money instead of losing it. Watch runs underneath all of it, passively, between sessions.

Four Agents, Each With One Job

Copilot — Baseline Engine

Learns a per-user behavioral baseline from day one: who they pay, when, how much, and which devices. A transfer that matches clears with zero added friction.

Shield — Six Signal Layers

Amount anomaly, new recipient, time-of-day, channel risk, narration keywords, and NIP response code. Scored and explained in plain language, inside a 200ms budget.

Ghost — Cooling Window

Routes an overridden block into a disposable holding account instead of letting the transfer clear outright, so even a fully-scammed user has a way back.

Watch — Passive Monitoring

Runs between sessions, watching for pattern drift and device changes, without adding latency to the transfer path itself.

One API Call, in Front of Every Transfer

Banks, fintechs, wallets, and gateways integrate with a single scoring endpoint. Context, device, and behavioral signals go in; a decision, a risk score, and a plain-language reason come back.

POST /v1/score
{
  "transfer_id": "txn_8f2ac1",
  "user_id": "usr_44210",
  "amount": 150000,
  "recipient": { "account_number": "0123456789", "bank_code": "058" },
  "channel": "app",
  "narration": "urgent help abeg"
}

{
  "decision": "FLAG",
  "risk_score": 78,
  "signals": ["new_recipient", "narration_keyword", "time_anomaly"],
  "action": "ghost_hold",
  "latency_ms": 142
}

Fast Enough to Sit in Front of Every Transfer

Not just the risky ones. The full pipeline has to clear before the transfer does, so it’s built to a hard latency budget, not a best-effort one.

<0ms

End-to-end response time, all four agents included

NIP's clearing window; Fable uses a fraction of it

Start Your Fraud Risk Assessment

See where your transfer flow is exposed. Get started with a free, 1:1 risk assessment for your institution.

Get Started