# Magentix AI > Magentix is a senior AI-governance practice that treats trust in autonomous systems as an engineering problem, not a policy aspiration. It builds the architecture that lets organisations prove what their AI and agentic systems actually did, under whose authority, at the moment they acted, and defend that record to boards, auditors, and regulators afterwards. Magentix exists on a single premise: as organisations delegate real decisions and actions to AI and agentic systems, the question that matters is no longer whether those systems are capable, but whether what they did can be understood, evidenced, and defended. The failure mode of AI adoption is rarely the dramatic one; it is the quiet, certain one of running autonomous systems no one can fully account for. Trust fails first, before capability and before compliance, because the evidence that would sustain it was never designed in. Magentix treats this as an architecture problem, solvable at the layer where systems act, not something to retrofit with policy documents after an incident. The practice is led by Martin Sansone, whose three decades in regulated financial services include standards architecture for the UK's retail payments authority (Pay.UK) and work spanning central-bank and government programmes - environments where evidence, not enthusiasm, is the currency. Magentix is not positioned as one more vendor in the AI-governance market, but as the practice defining the layer at which AI accountability is enforced, and setting the evidentiary standard others build toward. Its execution-assurance product, ARBITR, is built to become as standard for execution integrity as TLS became for transport security. ## Key concepts - **The Architecture**: Magentix's framework for accountable AI-native operation. Three layers decide whether an organisation's AI actions can be held to account - identity, memory, and execution. IdSolid holds identity and memory. ARBITR holds one part of execution, the evidence of what ran; carrying execution out and enforcing it belong to other systems, which Magentix builds its products to compose with rather than replace. - **ARBITR**: the evidence layer for execution. It records and proves what an autonomous system did, under whose authority, at the moment it acted, and renders that record as a board-grade evidence report legible to an auditor or a regulator without an engineer present to interpret it. ARBITR evidences execution; it does not carry it out or enforce it. ARBITR is in its pilot phase, selecting ten pilot organisations on lifetime-access partnership terms. - **IdSolid**: the sovereign-by-architecture layer for identity and memory. It holds each subject's identity attributes and AI memory inside their own cryptographically isolated Sovereign Pod, making data sovereignty a structural property of storage rather than a contractual promise. IdSolid runs an Early-Design-Partner programme of twenty places spanning both organisations and individuals. - **The AI Readiness Diagnostic**: a structured assessment that gives an organisation clarity on its AI position before it commits to deployment. - **automate.mu**: Magentix's Mauritius implementation team, delivering practical AI automation such as voice agents, invoice processing, and document summarisation once direction is set. ## Core pages - [Magentix AI home](https://magentix.ai/): the practice, the architectural thesis, and its three commercial lanes. - [The Trust Stack](https://magentix.ai/architecture): the architectural argument tying identity, memory, and execution together. - [The AI Readiness Diagnostic](https://magentix.ai/diagnostic): clarity before commitment, for organisations weighing AI adoption. - [About Martin Sansone](https://magentix.ai/about): the founder's background, the thesis behind the practice, and its current work. - [Insights](https://magentix.ai/insights): published thinking on AI governance, written as a discipline rather than a content category. - [Contact](https://magentix.ai/contact): how to begin a conversation with Magentix. ## ARBITR - execution assurance - [ARBITR overview](https://magentix.ai/arbitr): the execution-evidence thesis and what the product proves. - [ARBITR Pilot Programme](https://magentix.ai/arbitr/pilot-programme): ten pilot organisations on lifetime-access partnership terms. ## IdSolid - sovereign-by-architecture identity and memory - [IdSolid overview](https://magentix.ai/idsolid): the Sovereign Pod, and sovereignty as a structural property of storage. - [IdSolid Early-Design-Partner Programme](https://magentix.ai/idsolid/pilot-programme): twenty places spanning organisations and individuals. ## Insights - [Trust fails first](https://magentix.ai/insights/01-trust-fails-first): why trust, not capability or compliance, is the first thing to break in AI adoption. - [The diagnosis gap](https://magentix.ai/insights/02-the-diagnosis-gap): why organisations struggle to assess their own AI readiness honestly. - [What an AI audit actually is](https://magentix.ai/insights/03-what-an-ai-audit-actually-is): what genuine AI-accountability evidence looks like in practice. ## For agents - machine-readable interfaces Magentix publishes structured, machine-readable surfaces alongside these pages: - [ARBITR capability manifest](https://magentix.ai/arbitr/manifest.json): what ARBITR does and how to engage, as JSON. - [IdSolid capability manifest](https://magentix.ai/idsolid/manifest.json): the same, for IdSolid. - [Capabilities index](https://magentix.ai/.well-known/magentix-capabilities.json): a directory of these surfaces. - [Agent EOI protocol](https://magentix.ai/standards/agent-eoi/v0.1/): how an agent applies to a programme on behalf of an organisation, and how that application is verified. - Markdown versions of the core pages are available by appending `.md` to the path (for example https://magentix.ai/arbitr.md). A JSON intake adapter for agent expressions of interest is live, and its interactions are recorded as ARBITR Envelopes. MCP and A2A adapters are in development. ## Optional - [Privacy notice](https://magentix.ai/privacy): how Magentix handles personal data. - [Cookies](https://magentix.ai/cookies): Magentix uses only self-hosted, cookieless analytics and sets no tracking cookies.