Architecture and governance

Powerful agents, firm control

How eVitalyst is built so that AI agents can read, reason and act inside your ERP without weakening the controls your auditors and your board rely on.

Architecture

Five layers, one platform

Agents sit on top of the ERP core and act through it. The language model interprets and explains; the figures always come from the system of record.

Experience

Chat, executive overview, dashboards, advice inside forms, web and mobile

Agents

Business Assistant, Insight Developer, Entry Advisor, Data Entry, AP Invoice, Production Planning, MRP

Intelligence services

Language model reasoning, question-to-query over the ERP's data dictionary, retrieval and local vector memory (pgvector)

ERP core

Business rules, validations, document workflows, approvals, ledger, multi-company and multi-currency

Data

One PostgreSQL database for every company, with roles and access rights

Controls

Eight controls built into the design

Your permissions apply

Every question passes a role-based access check, and guardrails can be set for each role. Answers and actions stay within what the role allows.

Questions are read-only

Answering a question reads data; it never changes it.

Through the ERP, not around it

Agents use the same validations, document workflows and approvals as people.

Prepared, then approved

Agents prepare work; people approve what matters before it affects the books.

Every action recorded

Records keep who created and approved them, including work prepared by agents.

Company separation

Multi-company data stays separated by company and organisation.

Your choice of AI model

eVitalyst works with leading AI model providers, so the model can follow your policy.

Cost under control

Local intelligence reuses earlier work instead of calling the model every time.

Open standards

Connect your own AI tools through MCP

The Model Context Protocol (MCP) is an open standard for connecting AI assistants to business systems. With eVitalyst's MCP connector, assistants such as Claude or ChatGPT can look up and work with eVitalyst data under the same permissions as the user, with every call logged.

Also: REST and SOAP APIs

eVitalyst exposes its processes through web services, so agents and other systems use the same business rules as the application.

Integrations and APIs
Questions

Governance, answered

How does eVitalyst keep AI agents under control?

Agents act within the user's permissions, through the ERP's own validations and approval workflows, prepare work for people to approve, and every record keeps who created and approved it.

Can the AI change data when I ask a question?

No. Answering a question only reads data.

Which AI models does eVitalyst use?

eVitalyst works with leading AI model providers, so the choice of model can follow your company's policy.

What is MCP and does eVitalyst support it?

The Model Context Protocol is an open standard that lets AI assistants call tools in other systems. The eVitalyst MCP connector lets assistants such as Claude or ChatGPT work with eVitalyst under the user's permissions.

Free white paper

Agentic ERP: a guide for leadership teams

The five levels of agentic ERP, where value comes first, how to keep control, the economics of AI and a 90-day roadmap. Sent to your business email.

Get the white paper

Review the controls with our architects

Walk through security, permissions and governance with the team that builds eVitalyst.