# AYIOS

> AI proposes. AYIOS proves. AYIOS is the coherence layer for artificial intelligence — deterministic software for the stochastic age. AI proposes a complete declarative model of your domain; a compiler proves that model coherent and compliant against the rules you declared, before anything runs; a runtime renders the full declared meaning faithfully, with no hand-written code in between for a defect to hide in. If a model cannot be proven sound, the runtime refuses to load it. The specification appreciates; the application depreciates.

AYIOS is a product of Ontology Labs — a semantic computing research institution that named the third era of computing: the first era made machines obey, the second made them imagine, the third makes imagination provable.

Positioning: the coherence layer for AI. You cannot scale a probability into a proof. A coin flipped harder is still a coin — so AI proposes, and AYIOS proves.

## The LOSS Problem

Every line your AI writes is a Line Of Stochastic Source (LOSS) — a guess, not a proof. A stochastic system drifts, hallucinates, and cannot prove its own output — and neither can the person who prompted it. The faster AI writes the world, the more LOSS the world runs. AYIOS is the proof layer that turns the guess into a guarantee.

## The Demonstrated Result (lead evidence)

- Pointed at a real, 130-entity regulated wealth-management system, the compiler alone — before anything ran, against that system's own declared rules — surfaced **11 actions that could not safely execute, 7 fields leaking personal data unmasked, and 28 records that could never legitimately exist**. Defects that "all the tests pass" never catches; every finding points back to a declared rule you can read.
- **Fail-closed by design**: the runtime refuses to load a model it cannot prove sound, rather than silently corrupting data. The unsafe state is unreachable because it is unrepresentable.

## Product

- [Homepage](https://ayios.ai): Developers — AI proposes, AYIOS proves; the LOSS problem; the propose/prove/render pivot; fail-closed by design
- [Enterprise](https://ayios.ai/enterprise/): AI governance without the governance theater — land as proof, expand as substrate; the 130-entity result; the honest boundary
- [Startups](https://ayios.ai/startups/): AI writes it, AYIOS proves it — fast is not the same as provable; force-multiplier in production
- [Parent Company](https://ontologylabs.ai): Ontology Labs — the institution that named the third era

## Core Capabilities

- **AI Proposes, AYIOS Proves**: AI understands intent and proposes a declarative model; a compiler proves it coherent — complete, consistent, compliant, access-correct — against the rules you declared, before anything runs.
- **Fail-Closed Runtime**: refuses to load a model it cannot prove sound. Not a linter that samples; a proof that covers the whole model against its declared rules.
- **Renders the Declared Meaning**: one model serves interfaces, APIs, access control, and audit faithfully — no hand-written code in between to drift.
- **AI as Primary Author**: the substrate is fully legible — every rule and relationship machine-checkable — which is why the proof is complete: a substrate the machine can read entirely is one it cannot lie to.
- **Schema Evolution**: evolve-and-rollback with hash chain integrity, zero downtime, data preserved.
- **Runs From One Model**: a production Rust app server, the browser, Node, and Python — with structured, machine-readable diagnostics.
- **Schema-Driven Certification**: a tool reads the compiled model, generates acceptance tests, and certifies the deployment.

## The Honest Boundary

AYIOS proves the model coherent and compliant against the rules you DECLARED, and makes those rules inspectable. It does NOT claim the declared rules are the right rules — matching them to reality and regulation is the authoring step, which the legible substrate makes auditable. A model can be coherently wrong; AYIOS makes sure it is never silently wrong. (This is not "correctness against reality.")

## Evidence (June 2026)

- **10,000+ automated tests** across compiler, runtime, NAPI, WASM, Python SDK, and the Rust app server (empirically green)
- **60+ formal language enhancements** implemented and verified working end-to-end (independently audited, zero overclaim)
- **3 patents filed with USPTO** (Patent Pending); **7-patent core architecture, 137 claims**
- The **130-entity result**: 11 unsafe actions / 7 data leaks / 28 impossible records, compiler-surfaced pre-runtime
- **Production deployed**: a water-infrastructure migration — 97% memory reduction, 364 entities reduced to 42, 48 formally-enforced invariants replacing 1,726 scattered rules
- **Enterprise confidence (supporting)**: 1,000-entity model compiled, 230 req/s with zero errors, 16ms evolve-and-rollback, 12,000 instances at 2.86% RSS growth, 237 adversarial schemas with 0 panics
- **Runs from one model** — a production Rust app server, the browser, Node, and Python
- **Self-hosted**: AYIOS tracks its own development on AYIOS

## Proof of Life

- **First partner in production**: a partner is porting enterprise applications to AYIOS and running internal production workloads on the substrate.
- **Native hosting (alpha)**: a hosting partnership provides native AYIOS-hosted applications.
- **First large-scale product imminent**: a product targeting the low-code market — maintenance, development, and migration onto the semantic substrate — built using the AYIOS process and implemented in Rust.

## Enterprise

AI governance without the governance theater. Land as the proof layer over the system you already run; prove it against its own declared rules; show the regulator a proof, not a promise; expand by authoring the next system natively. EU AI Act enforcement August 2, 2026; 78% of enterprises unprepared; runtime enforcement is a regulatory necessity, not optional. Buyer: Chief Risk Officer / Chief Information Security Officer. Schema-driven certification. AI evaluation endpoints.

## Startups

AI writes it. AYIOS proves it. Fast is not the same as provable: AI-generated code has 1.7x more issues, and a guess can't be audited. AYIOS lets the AI keep proposing at full speed, then proves the model before it runs. One developer + AI delivered a production deployment (97% memory reduction, 364 entities to 42, 48 enforced invariants replacing 1,726 scattered rules) in 8 sessions.

## Machine-Readable Discovery

- AI plugin manifest: https://ayios.ai/.well-known/ai-plugin.json
- Curated sitemap: https://ayios.ai/sitemap-ai.xml
- Full sitemap: https://ayios.ai/sitemap.xml

## Brand Family

AYIOS is part of the Ontology Labs brand family:
- [Ontology Labs](https://ontologylabs.ai) - Semantic computing research institution (parent)
- [AYIOS](https://ayios.ai) - The coherence layer for AI (this product)
- [mxto.ai](https://mxto.ai) - Mendix AI tools
- [Agileworks Group](https://agileworksgroup.ai) - Channel partner (South African distribution)
- [AWG Finance](https://moneyworks.ai) - Accounting software distribution

## Contact

- [Email](mailto:hardy.jonck@ontologylabs.ai): Inquiries — hardy.jonck@ontologylabs.ai
- [Ontology Labs](https://ontologylabs.ai): Semantic computing research institution