One bitemporal ontology.One complex quant engine. One audit log.
Most "AI for finance" is an ungrounded chat window on top of third-party spreadsheets. Nyquist builds the full stack from first principles: a typed bitemporal ontology spanning 42 domains and 54 hubs, a deterministic quant core computing 34 derivative families, ALM, and systemic contagion, and autonomous agent desks that reason through live market data with explicit numerical kill-switches. Underneath: a lineage graph from every risk figure back to the tick and filing that produced it.
Vertically integrated
On-prem · by contract
Bitemporal · replayable
Vertical integration, not an AI wrapper.
Every layer is owned — from valid-time bitemporal state storage to stochastic PDE solvers and the agent desks on top. Market data and hosted model APIs are the only third-party dependencies, and every number an agent shows comes from the engine, not from the model.
★ Owned end-to-end — software-grade gross margins, high switching costs, and regulatory auditability.
Typed bitemporal ontology — 54 entity types, 42 domains, 54 hubs.
Not an unstructured data lake. Every object in Nyquist is a typed entity with explicit relational contracts and valid-time duality — a Trade links to its RFQ, Instrument, Counterparty, ISDA/CSA margin terms, Discount Curve, and Fund Book. Once a fund or bank maps its positions to this ontology, Nyquist becomes the permanent System of Record — high switching costs, and reproducibility a regulator can replay.
★ 54 entity types across 42 financial domains and 54 analytical hubs. Every fact stores [valid_time_start, valid_time_end] alongside transaction time for true historical reconstruction.
Time-travel any number back to its sources.
Designed around BCBS 239 risk-data-aggregation principles. Pick any figure in any committee report and unwind it across valid and transaction time to the exact tick and regulatory filing. Every run records its inputs and their digest, so the number can be recomputed as of that date.
Why incumbents can't pivot,
and wrappers can't compete.
Generic AI tools bolt chat prompts onto third-party spreadsheets: model-generated numbers, and customer data that leaves the perimeter. Legacy terminals run on 30-year-old relational databases that cannot support valid-time bitemporality or agentic swarms without breaking backwards compatibility. Four architectural pillars separate Nyquist:
Audit by architecture
Reproducibility isn't a post-hoc report — it's an inherent property of the bitemporal store. Every agent step, stress run, and Greek delta is replayable to its source. Built around BCBS 239 aggregation principles; compliance is assessed by the bank.
Agents are operators, not chat
Agents call typed mathematical primitives through an MCP server and act on real portfolio state. Outputs are structured committee memos with numerical kill-switches — never unstructured text you verify by hand.
Sovereign perimeter
On-premise or single-tenant VPC by contract. No model training on customer books. In-perimeter inference on compact domain models is on the roadmap, not in production.
One schema, many surfaces
Hedge funds, asset managers, and sovereign supervisors share the same ontology, quant engine, and audit log. Every feature built for one desk hardens the system for all — the codebase compounds.
| Architectural Dimension | Legacy Incumbents Terminals & risk platforms | AI Copilots & Wrappers LLM Prompts & Vector DBs | Nyquist Vertical Quant Platform |
|---|---|---|---|
| Mathematical Precision | Closed batch calculators | Model-generated numbers | Deterministic PDE & quant core (34 families) |
| Temporal Data Truth | Overwrite on update (no valid-time) | Stateless prompt window | Typed bitemporal ontology + lineage graph |
| Regulatory auditability | Days of manual spreadsheet reconciliations | Untraceable sources | Built around BCBS 239 aggregation principles; compliance assessed by the bank |
| Autonomous Desk Swarm | Zero agent capability (human manual) | Conversational chat / text summary | 36 dialectic agents with numerical kill-switches |
| Deployment Boundary | Multi-million legacy on-prem lock-in | Public cloud (customer data egress) | On-prem or single-tenant VPC by contract |
| Time-to-Value & Commercials | 6–18 month procurement · $500K+ floor | Self-serve toy (not procurement-grade) | 60-day live book pilot · 100% credited to license |
Procurement-grade from day one.
Built for Tier-1 capital markets, regional banks, and systemic supervisors. Runs inside your perimeter by contract, with no model training on customer books.
See the platform run on a live book.
The full platform runs in your browser — you enter your own positions. Watch stochastic pricing, stress tests, bitemporal valid-time lineage, and autonomous research desks converge. No calls, no scheduling.