Broker-dealers & lenders
Margin, risk and agents behind your own firewall, under your brand.
Agents read the market and propose a trade. A margin firewall checks it against the rules. A named person approves it. Every step is signed into a ledger you can audit.
Built for broker-dealers, lenders, funds, advisors and independent traders. It runs on your workstation, your firm's server or your own GPU cluster, with no vendor cloud in the middle.
The same four steps sit under every use case, from a single trader's desk to a broker-dealer's data center.
Each role has a worked example in the field guide and a live desk in the playground.
Margin, risk and agents behind your own firewall, under your brand.
A working system to lead with, and models trained on hardware you own.
A desk on your own machine. Client records stay where they already live.
Public services are reached through a Cloudflare Tunnel. Each card says what it is today, not what it will be.
Agentic trade intelligence
Pre-trade margin firewall for AI agents
Order management for agents
Independent SPAN 2-style futures margin
Cboe XSPBW / QSB analytics
Tax-aware, multi-account rebalancer
Single-stock leveraged ETF analytics
Point of No Return margin survival
ieBalancer rebalances one account, a household or a batch of accounts, and applies the tax rules you set with your tax advisor to every sell it proposes. The order set is validated at your broker-dealer, a person approves it, and the broker-dealer's book is read back as the record.
ieBalancer applies tax rules; it does not give tax advice. Wash-sale and basis figures are estimates, and your broker-dealer's 1099-B is authoritative. In development: the rules below are built and tested on a simulator, and harvesting as a scheduled run is next.
What changes is the box, never who owns it.
No vendor cloud in the middle at any size. Client PII stays where it already lives.
Machines can discover, call and pay for the services without a human in the middle of the plumbing.
Every service publishes /llms.txt and /llms-full.txt: endpoints, schemas and capabilities in plain text.
Model Context Protocol endpoints let agents find and call tools with typed schemas and structured results.
Spark · Mercurius · SPAN 2Paid endpoints answer 402 and accept USDC on Base Sepolia testnet. No API key to manage.
Spark · MercuriusOne broker abstraction. Orders route to the broker-dealer you already use; paper accounts and a simulator today.
OMS · SSETFGrace-Blackwell GB10 nodes serving open-weight models through TensorRT-LLM and vLLM, with NVFP4 and FP8 quantization.
GPU clusterProposal, check, approval and fill share one trail. The ledger is hash-chained, so a changed record breaks the chain.
platform-wideA five-agent SPX 0DTE options pipeline on our own Grace-Blackwell nodes. Local models reason inside a deterministic risk cage, and a person approves by default. It trades in a broker-dealer paper environment today, with no client capital.
A structured fund program is forming around it: legal formation, compliance and independent service providers. Investor materials are shared only with qualified investors, under NDA.
A senior engineering leader with 20+ years across institutional trading infrastructure and frontier AI systems. I build end to end: the LLM agents, the risk math and the GPU metal they run on.
Currently leading the AI-native modernization of Organization's Risk Monitor, while building UnifiedQuant, Mercurius, the Agentic OMS, UnifiedRBM and LocalInfra, a Grace-Blackwell cluster that runs 235B-parameter inference.
Forty-five minutes. We walk the platform live, you name the market and constraints, and we leave you with a scoped pilot. Nothing is installed and no client data is shared on the call.