Investors
The category nobody occupies yet.
Every agent vendor ships a control plane for its own agents. Nobody records, controls and reconciles what all of them did, in business terms, for the person who has to sign. That gap is what Meru Assure is built for, and vertical platforms are how we reach it.
This page carries no financials, no customer names and no traction numbers. It argues the category and the design instead.
The category
Four camps. Each one borders the gap. None of them is in it.
Agent accountability is being sold four ways today, to four buyers, and each of them solves a different problem well. Between them sits the question a controller actually asks: what did every agent in this company do to its systems last quarter, and can I show somebody.
A map with four dashed boxes at the corners and one solid dark box across the middle. The four corners are agent security, sold to the chief information security officer; vendor control planes, covering one estate each; AI governance and GRC, which produces policy documents; and observability, which produces developer traces. A dotted line runs inward from each corner towards the middle box — Meru Assure: business controls, evidence and reconciliation across every vendor’s agents — and each line stops short of it rather than entering.
Who covers what today, and where each one stops.
| Camp | What it sells | Who buys it | Where it stops |
|---|---|---|---|
| Agent security | Guardrails, prompt defence, data-loss prevention around the model | The chief information security officer | It stops the bad request. It says nothing about whether the approved action was carried out correctly. |
| Vendor control planes | Policy, logs and approvals for that vendor’s own agents | The platform owner who bought those agents | One estate each. Four vendors means four control planes and no single answer. |
| AI governance and GRC | Frameworks, registers, risk assessments and policy documents | Risk and compliance | It describes the controls. It is not wired to the action, so it cannot show one was enforced. |
| Observability | Traces, spans, token counts, latency, error rates | The engineering team | It proves the call happened. It does not prove the business change happened, or was allowed. |
| Meru Assure | Business controls, bound approvals, reconciliation and evidence packs across every vendor’s agents | The controller, internal audit and risk | The gap the other four border. Built to be bought without Meru Run, and to record agents we did not build. |
Why now
The capability arrived first. The accountability did not.
Agents became able to act on real systems faster than anyone built the controls, the evidence and the reconciliation around them. The gap, not the model, is now the thing that stops a deployment.
-
Prediction
Over 40% cancelled by the end of 2027
Gartner names three reasons agentic AI projects are expected to be cancelled: cost, unclear value and inadequate risk controls. Two of those three are accountability problems wearing other clothes.
Gartner, 2025 -
Regulation
The rules ask for governance you can show
The EU AI Act, ISO/IEC 42001 and the NIST AI Risk Management Framework all ask an organisation to demonstrate how an automated decision was governed. A chat transcript shows a conversation. It does not show a control.
-
Insurance
Somebody has to price this risk
An underwriter looking at an autonomous system has to ask what it was allowed to do, what it actually did, and what happens when it is wrong. A policy document does not answer that. A record of every consequential action does.
What is built
One small Core. Two products on it. One vertical platform above.
The shape matters more than any single feature. Everything above Core inherits the record, and everything below Core is a plug point, so a new model or a new system is an adapter rather than a rebuild.
The four layers, and why each one matters to a company rather than to an engineer.
| Layer | What it is | Why it matters here |
|---|---|---|
| Meru Core | Identity and policy, a capability registry where every action declares what it touches, durable execution, a model gateway, and an append-only, hash-chained evidence ledger. | Deliberately small. The record is a property of the platform, not a feature somebody switches on. |
| Meru Run | Agents that complete bounded work inside a customer’s own systems under an operating contract, with packs supplying the workflows and policy defaults per department and industry. | Where the work happens, and where usage grows with the number of tasks handed over. |
| Meru Assure | Business controls on consequential actions, approvals bound to exact parameters, reconciliation against the system of record, and evidence packs with a read-only auditor view. | A separate buyer, a separate budget, and it can be sold to companies running somebody else’s agents. |
| JhiOn | Longevity & Health Span Management: the first vertical platform built AI-native on Ayion Meru. | The vertical that forced the Core to exist, and the template for the ones that follow. |
What this page does not claim
Ayion has no customers to name, no certifications, no audited evidence pack and no results to report. When any of that becomes true it will appear here in plain words, with the date. Until then this page argues the category and the design, which are the parts we can actually defend.
The thesis
Four bets, stated plainly enough to be wrong.
Each of these is a decision that closes doors. Taken together they are the reason the platform looks the way it does.
-
The doer cannot be the witness
An agent vendor recording its own agents asks an auditor to trust the doer’s own log. Assure records what any agent did, Run cannot write to the ledger, keys are customer-held, and a third party can verify a record without us. That rule is the only reason one company can honestly sell both.
How Assure works -
Evidence outlives the model
The models will be replaced every few months, and each replacement resets whatever advantage sat in the prompt. What was allowed, who approved it, and whether the system of record agreed does not reset. The durable asset is the ledger.
-
Vertical first, then the Core
Build for one industry’s problem in that industry’s language, then push the reusable parts down into the platform. It is slower than selling a horizontal tool to everyone, and it is the only way the controls end up specific enough for anyone to sign.
The first vertical -
Small Core, open edges
Models, agents, systems, controls, packs and channels all plug in. A company running four vendors’ agents does not want a fifth estate; it wants one place that covers all four. That is only possible if the Core stays small.
Read the architecture
Request the deck.
Tell us your firm and your stage and we will send it. It goes further into the architecture, the category and where the build actually stands.