We build things and we advise on things, and we know the difference between the two. Most engagements start with an institution that needs one and discovers it needs both. We work with a limited number of institutions at a time.
Browser-only calculators, policy engines and agent workflows built to your rules. Each tool computes on the user's device, keeps no personal data, and returns results an auditor can recompute. This is how we built AINumbers and ApexLogics, more than 1,200 tools and workflows between them.
A property-test floor on every live kernel, exhaustive enumeration and property testing on named kernels, a machine-checked Dafny and Z3 proof, and contract checks across every workflow. Known debt is published as shrink-only baselines. How it works ↗
The Dafny proof of a Reg Z Appendix J APR kernel showed it compounds interest where the regulation prescribes simple interest, while it still matched all four of the regulator's worked examples. We disclosed it and track the repair. The finding ↗
Our workflows follow OpenChainGraph, the open specification we wrote for hash-linked agent decisions, now at v0.8.13 after a series of public releases. Every result carries a SHA-256 execution hash anyone can recompute. Read the spec ↗
Publish your products to AI assistants through an MCP server like ours. We operate one of the largest public MCP servers in finance: more than 730 tools on one endpoint, listed in the official MCP registry as co.ainumbers/tools. We know what agents need from a tool list and what breaks at scale.
More than 670 deterministic nodes carry a risc0 Groth16 zero-knowledge proof of execution. New nodes are usually proved within a few days to a week of going live. AINumbers ↗
Besides AINumbers, ApexLogics runs its own MCP server with a hash-verification tool, and the Omega Centauri Society runs an 11-tool science server. OCS server ↗
Our browser pages register the same tools through WebMCP, so an agent running in the browser can call them on the page itself, with nothing sent to a server.
We draft, check and publish technical research with AI assistance, disclosed on every paper. The derivation and verification habits we apply to code carry over to the literature.
While implementing the formalism of Prager et al. (2017, ApJ 845:148) for an Omega Centauri study, we found its jerk noise-floor constant, printed as 3.04 and adopted by later work, evaluates to 3.4596. Five independent evaluations and two Monte Carlo simulations reject the printed value. Published as a preprint. The erratum ↗
The Omega Centauri Society has published ten working-draft preprints, from a pre-registered analysis of the cluster's central dark mass to a residual census of its X-ray sources. The papers ↗
Objective, openly run benchmarks for language models and MCP servers go live in the coming days.
For commercial banks, central banks and large corporates, with a primary focus on Latin America, Africa, the Caribbean and South Asia. We have advised on and supported production deployments across frontier markets.
Design and integration of tokenized account-to-account rails for B2B and B2C cross-border transactions and remittances. See the A2A payments guide.
Issuance architecture, reserve management structure and regulatory positioning for commercial and central banks, fitted to institutional context and jurisdiction.
Central bank digital currency design and deployment: interoperability with existing payment systems, integration with private-sector infrastructure, and the structural questions central bank teams face.
Standardized tokenized MMFs in local currency. Tokenized USD funds already exist; the gap is local-currency equivalents for non-dollarized economies, which banks can offer corporate clients as a differentiated product.
High remittance volumes, regulatory openness in key markets, and banking sectors modernizing cross-border infrastructure.
Mobile money penetration and CBDC experimentation, especially for corridors that are expensive and slow today.
Direct experience of the region's regulation, correspondent banking constraints and remittance dynamics.
Remittance volumes, regulatory structure and a commercial banking sector with ongoing opportunity as tokenized infrastructure matures.
The team has advised and built products across more than 20 institutional blockchain projects, from layer-1 infrastructure to DeFi, enterprise ledgers, agentic AI trading and tokenized finance. Some engagements were advisory and research; others were direct product development and deployment. Client names are not listed; get in touch to discuss experience relevant to a mandate.
Market design, market making, resolution mechanisms, and economic forecast markets treated as financial instruments. Example: Canonicals ↗
Autonomous trading agents for decentralized protocols: factor research, multi-factor strategies, realistic backtesting and non-custodial execution on perpetuals and prediction markets. Example: BanZhuan ↗
Design, issuance, reserve structure and regulatory strategy, including stability mechanisms for algorithmic and collateralized models. Example: Auton ↗
Public and permissioned layer-1 projects: consensus design, tokenomics, validator economics and go-to-market positioning.
Interbank settlement, trade finance and cross-border payments on permissioned platforms. See the enterprise blockchain briefing and A2A on enterprise ledgers.
From early development to institutional integration: liquidity mechanisms, governance and risk models, and AMM fee structures, capital efficiency and impermanent loss.
Onchain perpetual protocols, margin systems, oracle design and liquidation mechanics, and how they hold up in regulated contexts.
NFT infrastructure for digital IP, onchain provenance and royalties, plus token economies and ownership structures for decentralized networks.
Verifiable credentials and self-sovereign identity, plus blockchain analytics for bank compliance teams: transaction tracing, entity identification and AML tooling.
Tell us what you are evaluating and we will say plainly whether we can help.