Two layers of intelligence. One governed stack.
Proprietary detection models built by quant researchers find the signal. Frontier AI — Claude and GPT over MCP — multiplies what you can do with it. Both run under your governance, in your dedicated environment.
The moat: proprietary quant/ML detection
Designed and built in-house for one domain — trading operations. This layer is what watches the book.
Toxic flow
Classification of flow that extracts value from your pricing, your latency or your rules — latency and stale-quote arbitrage, news-window exploitation, gap and rollover strategies — scored as it develops, with evidence attached.
Abuse & fraud
Coordinated behaviour across accounts, multi-accounting signals, opposite-side hedging across entities, bonus and evaluation gaming — the patterns designed to stay just inside the letter of a rulebook.
Anomalies
Statistical departures that don’t match a named pattern yet — surfaced early, so the first time you see a new strategy isn’t in the monthly P&L review.
Layer two — frontier AI, governed
Claude (Anthropic) and GPT (OpenAI) — including GPT-5.5 in-app — over MCP, the open standard. The governance is drawn in, not promised.
Natural-language querying
Interrogate operational data in plain language through governed MCP tools — accounts near limits, cohort behaviour, flow profiles.
Automated narratives
Plain-language explanations of what the detection layer flagged and why; daily summaries; case write-ups drafted from the evidence chain.
Task automation
Governed automation of the repetitive middle — enrichment, first-pass review, report assembly — with human checkpoints where actions have consequences.
In-product copilot
An operator copilot grounded in your data and policies, scoped to what each role may see and do. OAuth 2.1 · RBAC · full audit.
Claude and Anthropic, GPT and OpenAI, and MCP are trademarks of their respective owners. Their use here indicates technical integration only and does not imply endorsement or partnership.
The off switch is part of the architecture
How frontier models coexist with your data, your servers — deliberately, and under contract.
- Enterprise API terms. No training on client data; zero or strictly limited retention — contractual, not aspirational.
- Scoped context only. The dataset never leaves your environment; governed MCP tools carry minimal context to the model.
- Everything logged. Which role invoked what, with which data scope, and what came back — AI access is part of your audit trail.
- Disable per client. For ultra-sensitive deployments the external AI layer switches off entirely — Layer 01 detection keeps running inside your environment.
Intelligence you own the return on.
Ask us the hard governance questions in the demo — retention terms, tool scoping, the off switch. That conversation is the product working as intended.