Deterministic fintech tools that never touch a server
AINumbers.co is a working catalog of browser-based calculators and checkers for people who run payments, compliance, treasury and risk day to day. Every tool runs on your device: same input, same output, nothing leaves the browser tab.
A working reference for payments practitioners
Most fintech work happens in spreadsheets, vendor portals, and half-remembered rulebooks. AINumbers.co replaces the repetitive parts with focused, single-purpose tools: an ISO 20022 message builder, a FedNow participation scorer, a Basel RWA calculator, a sanctions-screening batch checker, a DORA gap analyser, and a few hundred more.
Each one does a single job well: you give it structured inputs, it returns a defensible, reproducible result, and where it matters it exports a machine-readable mandate or report you can hand to a counterparty, auditor, or your own systems. There are no accounts, no paywalls, and no "request a demo" gates between you and the answer.
A separate formal-verification pilot goes further on four named kernels, each carrying a formal claim tied to a human-signed spec and cited on Methods by the tier that applies: exhaustive enumeration, property-tested over stated ranges, or a Dafny machine-checked proof. See Methods for what is and is not covered.
Four principles with no exceptions
connect-src 'none', so the page cannot make a network call even if it tried. Every tool also works fully offline: disconnect the network and it runs the same. You can safely work with sensitive structure (just not real personal data).Think of AINumbers as an inverse database
AINumbers keeps no database of your inputs. The phrase inverse database is a metaphor for the shape of the thing. A typical hosted app asks you to log in, sends your data to its servers and keeps the record there. Here each tool is a public page, your data stays in the tab, and the record leaves with you.
The scenes above describe the browser tools, and the privacy line below describes those same tools. The MCP server at mcp.ainumbers.co is a separate surface that, by default, computes server-side for nodes with a registered kernel that need no GPU, stores no inputs, and records pseudonymous call metadata (tool name, pseudonymous caller hash, success/failure, latency, chain depth). See how the server treats your data.
Two related things doing two different jobs
AINumbers.co and the demo suite at PostOakLabs.com are easy to confuse, so here's the distinction in one line: AINumbers is a self-serve reference you operate yourself; the demo suite shows those same capabilities running end-to-end.
- Open any tool, enter your own inputs, get a result
- Grouped by category and hub, or sequenced by goal-based tool chains
- Discovery aid: find the right tool and the right order
- Free, client-side, zero PII, no sign-up
- Pre-filled inputs and commentary across a full scenario
- Organized by archetype: Agentic Runtime, RegTech, BaaS, Processors, Stablecoin Issuer
- Demonstration: shows what the tools can do together
- The consulting context behind the catalog
Use AINumbers when you want to do the work yourself. Visit the demo suite ↗ when you want to see a complete workflow run before you build it.
The tools compose on purpose
The catalog isn't a pile of unrelated calculators. Tools share input and output conventions so the result of one feeds cleanly into the next, and the site exposes that structure three ways. Categories group tools by domain. Integration hubs gather everything you need for a single area, like open banking, DORA resilience, AML/KYC, B2B payments, and dozens more. And tool chains organize sequences by goal rather than category: "validate an ISO 20022 cutover," "build a DORA submission," "design a BaaS programme."
That intentional composition is the point. Whether you're a practitioner finding the next step or someone evaluating the catalog as a whole, the tools are meant to be read as a coherent system: each piece useful alone, and more useful in sequence.
Built for machines as much as people. Because every tool is plain, deterministic HTML, it reads as clearly to code as it does to a person. Many ship an MCP-ready manifest.json and a Policy Mandate export, so agents can call them inside automated pipelines. Results are anchored under the open OpenChainGraph standard with an execution_hash that any party can recompute and check, and the suite runs a live MCP server at mcp.ainumbers.co/mcp. The site also publishes an llms.txt with clean semantic markup so crawlers and language models can index it reliably. The whole catalog is open source. Fork it on GitHub ↗.
Standard candles for machine reasoning
In astronomy, a standard candle is an object whose true brightness is known, so astronomers can measure everything around it by comparison. Tools on AINumbers.co are built to work the same way for machine reasoning. Each one is deterministic, runs with no network calls, and carries a hash-anchored result, so the same inputs always give the same output and anyone can re-verify it later. That property gets more valuable as AI output grows more fluent and harder to check by eye.
Rather than trying to cover every problem across fintech, regtech and insurtech, the goal is simpler: give a model, and the person relying on it, enough fixed reference points to calibrate the larger workflow it assembles. The ChainGraphs are the rungs of that ladder, where each hash-anchored handoff is one checked step from a known result to the next.
If you're weighing whether this approach is genuinely new (and why it only became buildable recently), we lay out the honest prior art (in-toto, SLSA, OpenLineage, AP2), the convergence that enabled it, and the caveats on Why OpenChainGraph →.
Built and maintained by Post Oak Labs
It is a payments and fintech consultancy, and AINumbers.co is its open, public reference library: the deterministic tools behind the consulting work, released free under CC BY 4.0 so anyone can use it, fork it, or build on it.
Spotted a wrong result, want a tool that doesn't exist yet, or have a payment code we're missing? Get in touch; every message gets read.