Enterprise applications your AI agents can work in safely.
The Zengentic™ platform gives every application identity, permissions, segregation of duties and a full audit trail from the first day. An AI agent can do only what the person it works for may do.
- Action
- Quantity amended, 50 to 75 MW
- For
- A. Trader
- By
- AI assistant, acting for A. Trader
- Allowed
- Amend trades: held by both
- When
- 1 Oct 2026, 14:32:07 UTC
- Next step
- Awaiting approval by Middle Office
AI agents are arriving in systems built for people.
Your control framework assumes every actor is a named person. When an AI agent uses a shared account or a person's login, that assumption no longer holds. How agents are governed
Built on governance, ownership and trust.
People use the web application. Services use the REST API. AI agents use MCP tools. Every request passes the same permission check, whichever door it uses, so the controls you own apply to all three.
What a person cannot do on screen, their agent cannot do through a tool. The platform enforces this in its structure, not in a policy document.
How the Zengentic platform works- Identity and groups
- Segregation of duties
- Audit trail
- Lifecycles and approvals
- Document templates
- Task scheduler
- Configuration
- C# and Python extensions
- Secrets in Key Vault
Everything an application needs before the first line of business logic.
A new application brings its own data, rules and screens. The rest comes with the platform.
API-first and MCP-first
Business operations are available as API calls and as AI agent tools, through the same permission check as the screens.
Agents act for a named person
Each request carries two identities: the person it is for, and the agent at the controls.
Enterprise identity
Microsoft Entra ID, Windows Active Directory and API keys, with one permission model.
Four-eyes approvals
Any step can need its own permission. The person who books can be required to differ from the person who approves.
Customise without changing the core
Fields, screens, objects and rules as configuration. Code in C# or Python at declared points.
Auditable, not merely logged
Trades, prices, documents and configuration keep every earlier version, with who changed it and for whom.
- Microsoft Entra ID
- Active Directory
- Azure Key Vault
- SQL Server
- Claude, via MCP
Building the system becomes cheap. Keeping it auditable does not.
AI coding tools make enterprise applications faster to build. Identity, approvals and an audit trail still take careful design.
The Zengentic platform provides them from the start, so your team, and the AI agents that help it, build on top of them.
Already running: ZenCTRM.
Commodity trading and risk management, from trade capture to invoice, built on the Zengentic platform. AI agents book, value and settle trades under the same four-eyes approvals as the desk.
Visit zenctrm.comPhysical and financial
Power, gas, oils, agriculturals, metals, emissions and derivatives.
Reproducible valuations
An end-of-day run can be re-computed exactly as it was.
Settlement and invoicing
Cash flows, netting, credit notes and a journal for your ledger.
We build with you, then hand it over.
Zengentic works directly with your team. We design, build and run governed applications on the Zengentic platform, and connect your AI assistants to the systems you have.
Shape
We map the process, the roles and the approvals. You see working screens early, not a long specification.
Build
AI agents do much of the building, through the same governed tools. A person approves every change.
Run
On-premises, cloud or hybrid. Your data stays in your database.
20 years in commodity systems.
We have implemented, tested, migrated and run trading and risk systems for energy, oil, agriculture and metals firms across Europe and Asia. Now we build the platform we wanted then.
See consulting servicesQuestions we are often asked.
Who is the Zengentic platform for?
Organisations that need a governed system of record and want AI agents to work in it. Our first application, ZenCTRM, is for commodity trading firms.
Where does it run?
On your servers, in your cloud, or both. It runs on Windows Server with SQL Server. Your data stays in your own database.
Do we have to use AI agents?
No. Everything works through the screens without an AI assistant. Agents are optional, and an administrator decides which people and agents may use them.
What information reaches an AI provider?
Only through an assistant a person chooses to use. When someone works through an AI assistant such as Claude, their messages and the results of the tools the assistant calls go to that assistant’s provider, and the tools return only what that person may see. the Zengentic platform itself does not send your data to an AI provider. An administrator decides which assistants and people may connect, and any model keys are held in your own Azure Key Vault.
How does implementation work?
We start with a short workshop on your processes, approvals and reports. We configure first, and write code only where it is needed. Your team tests in its own environment before go-live.
Can we change it ourselves?
Yes. Administrators change screens, fields, lifecycles and approval steps without a release. Developers add extensions in C# or Python. Every extension is checked for compatibility before an upgrade.
Do you work on other trading systems?
Yes. Our consulting covers implementation, testing, migration and integration on the system you already have.
See an AI agent work under your controls.
In 30 minutes we show an AI assistant booking work in a live application, a person approving it, and the full record of both. Bring a real process.