Platform & Security Architecture | Xpertly
Platform

The architecture your security team will actually approve

AI agents act on your real systems through one governed plane — deterministic permissions, full audit, deployed in your environment. The model proposes. The plane decides what happens next.

The trust chain

Nothing that reasons touches your systems

A client, a channel, or an event reaches the plane; the plane checks what's permitted; a control agent executes deterministically; every action is logged. Trust sits in the permission system, not in the model.

CLIENT · CHANNEL · EVENT

Claude/Copilot · voice/web · SIEM/monitors — any initiator. It proposes; it never touches your systems.

Xpertlythe governed plane
Authenticate & enforce permissions

OAuth via your IdP · deterministic, per-user grants

Agent executes

Deterministic flows & command sets — the only things that touch your systems

Assets

Your systems, files, commands & knowledge

EVERY ACTION AUDITED — who or what asked · through what · did what

Most of the industry calls a model-with-tool-access an "agent." We don't. An Xpertly agent is a governed unit, composed inside the plane:

The mind

The model, arriving through any MCP client (Claude, Copilot, Cursor, your own apps): reasons and proposes. Never executes.

The hands

Deterministic flows and governed command sets: the only things that touch your systems.

The knowledge

Documents, and the living ontology of your business: a graph of systems, relationships and rules, scoped to what this agent, for this user, is permitted to know. It deepens with use and can't be copied — the second moat.

The conscience

The grants, policy, and audit trail the plane enforces on every action.

An Xpertly agent can't exist ungoverned — take away the plane and you don't have a weaker agent, you have no agent at all. That's the difference between governed by construction and governed by wrapper.

Two substrates, one plane

How agents act, and how they understand

Every Xpertly agent draws on two foundations. Flows are how agents act — deterministic operations that run exactly as defined, and real-time retrieval when you need to know what's true right now. The ontology is how agents understand — a living graph of your business that answers what's related and why. Ask "is my order shipped?" and a flow retrieves it live. Ask "why did this go wrong?" and the graph traverses the relationships. Xpertly routes every request to the right path automatically — and having both, with the intelligence to choose, is something flow-only and graph-only systems can't match.

Core capabilities

Built for the security review, not around it

Deterministic execution

Flows run exactly as defined, or not at all. Same input, same governed action, every time.

Permissions below the model

Grant each user specific agents, each agent specific assets. A jailbroken model can only ask; it can't act outside its grants.

Actions and knowledge, both governed

Control what an agent can do and what it can know, per user, per agent.

Any client, channel or event

Claude, Copilot, a voice call, or a SIEM alert — all reach the same plane. Governance attaches to the user and agent, never the client.

Your environment

Cloud, on-premises, or sovereign. The model and data never leave your walls. Integrates with your IdP (Okta, Entra, others).

Audit-grade by default

Every action, executed or blocked: who or what asked, through what, did what. SIEM-exportable.

Technical teams may know this layer as an AI gateway or access broker. The difference: ours owns the execution, not just the access.

The living ontology

It understands your business — and that understanding can't be copied

Xpertly maps your operation into an ontology: the systems, the relationships, the rules. It's unique to you, it deepens with every workflow, and it can't be carried out the door. A competitor can rebuild the software in a quarter — they can't rebuild three years of your operational understanding. That's why an agent here doesn't just fetch a number; it knows that complaints spiked because a firmware update hit 47 sites, three with tickets still open from last month. The same ontology powers fraud detection that reasons across relationships, voice agents that know a caller's full context, and insight you can ask for in plain language. Governed execution gets AI into your business safely. The ontology is why it gets better the longer it's there.

Deployment

Built for production, deployed on your terms

In your environment

Cloud, on-premises, or air-gapped/sovereign.

Your identity fabric

Integrates with your IdP; agents and users carry distinct identities.

Configured, not coded

Agents and flows defined without writing code.

Built for production

Encrypted and authenticated throughout; isolation per deployment.

Answer "is it safe?" in four weeks.