The first OTC derivatives platform where AI agents don't just assist — they participate. Every module discoverable. Every output explainable. Every change safe.
See how ↓
Legacy trading systems were designed before the internet. They produce answers but not explanations, resist automation, and make change dangerous.
A P&L number moves overnight. To find out why, someone opens a spreadsheet and spends the morning reverse-engineering the calculation. The system produced the number. It cannot explain it.
every desk, every morningA new product type requires a quarter of development because everything is coupled. One pricing change triggers a cascade of regression risk across unrelated products.
one change, a hundred regressionsChatbots bolted onto systems built in the 1990s. AI cannot reason about data it cannot access, types it cannot interpret, or rules buried in procedural code.
bolted on, not built inEvery module exposes discoverable interfaces. Every data object carries its own context. Agents don't scrape the UI — they call the same APIs the system calls itself.
Describe a trade in plain English. The system structures, validates, and books it. So a trader spends seconds on entry instead of minutes fighting a form.
Business rules encoded in the type system. Malformed data cannot be constructed — not just detected downstream. So bad trades are impossible to book, not just caught later.
Every module exposes read capabilities via Model Context Protocol. So AI agents can query your book, explain a risk number, or draft a trade — without a human translating.
Every output carries its full provenance chain. When a number looks wrong, diagnosis takes minutes — not hours in a spreadsheet.
Every function that transforms data takes a LineageContext and returns one. Provenance is structural, not bolted on. So when a number changes, you trace it to the input that moved — in seconds.
Every output declares how it was produced: deterministic, heuristic, or AI-assisted. So you know which numbers to trust for margin calls and which to double-check.
Replay any subset of events, target any module, produce results into a shadow projection. So reconciliation, debugging, and regulatory reconstruction are the same operation — not three separate processes.
A modular monolith with event-driven communication. Every module is independently extractable. Every event is immutable. Every change is safe.
The event store is the authoritative record. Materialised views are projections rebuilt at any time. So "what was the state at 3pm Tuesday?" is a query, not a forensic exercise.
Changes to one product type cannot break another. Separate pricing, validation, and risk per product. So a new swaption model ships without touching your IRS pricing.
Modules communicate through the event bus and MCP. When a module needs independent scaling, swap the transport. So the valuation engine scales to production load without a rewrite.
Business invariants encoded as machine-checkable property tests. They catch what unit tests miss. So "the code works but the logic is wrong" stops happening silently.
Scion covers the full OTC derivatives lifecycle — rates, credit, FX, equity, commodity — from trade entry through valuation, risk, and P&L attribution.
Launch the app