Maerak is an enterprise decision intelligence platform. It turns the data in your ERP and business systems into recommendations, approvals, commands and verified outcomes — over a declared enterprise ontology, durable workflows and AI agents.
A request flows down to the decision layer and stops there. Your systems of record stay where they are and stay in charge.
A planner workspace built with React and Fluent UI, where people review and approve recommendations, with Power BI for analytics.
Foundry-served AI agents, each with its own Entra identity, passing the same authorization checks as a person.
Durable orchestrations on the Durable Task Scheduler. Every change runs inside a workflow, which can wait for a person's approval before it goes on.
Recommendation, approval and command services, and an append-only decision ledger that records the recommendation, its evidence, the approval, the command and the outcome.
An enterprise ontology kept in git: a shared core and one overlay per tenant, compiled into a catalog that agents and services read.
Each tenant's own Microsoft Fabric workspace and warehouse, with raw data landed and modelled into stored marts, and semantic models over the same grain.
System-of-record data lands in the tenant warehouse. Commands travel on Azure Service Bus through a transactional outbox, with dead-lettering built in.
Each of these has its own source prefix in the tenant warehouse, so every table says which system it came from.
Maerak owns the analysis and the decision. It never owns your inventory, orders or ledgers.
Inventory, purchase orders, vendors, assets and financial transactions stay in the systems that own them today.
The platform analyses, recommends, approves and issues commands. It does not post transactions.
Agents reason over declared business types and their links, not over raw tables.
A change happens only inside a workflow, and only through the command service.
Who acted, why, when and on what evidence is written to the decision ledger, which is append-only.
An agent can recommend, but it never approves its own recommendation.
One loop, recorded at every step.
Maerak reads the tenant's stored warehouse data through the ontology and proposes a change, with the evidence behind it.
A person approves or rejects the recommendation under their own sign-in, and the ledger records the decision.
The approved change becomes a command. A person carries it out in the system of record, which is the only way a system of record changes.
The next load from the system of record checks that the change happened, and the outcome is written to the ledger.
Tenancy is part of the design, not an afterthought. Each layer says how it keeps one tenant apart from another.
Each tenant has its own Fabric workspace and warehouse. No workspace is shared between tenants.
One Azure SQL database per tenant, with the tenant's identifier on every row.
A tenant's database mirrors only into that tenant's own Fabric workspace.
Every message carries the tenant's identifier, with a separate Service Bus namespace where isolation calls for one.
What an agent remembers is kept per tenant.
One shared core, and an overlay for each tenant that binds the core's types to that tenant's data.
One codebase for both clouds.
Maerak is built for commercial Azure deployments and Azure Government deployments from the same code.
Nothing in the platform may depend on a capability that Azure Government does not have.
Every component in the stack records whether it is available in Azure Government, where that was measured, or that it has not been measured yet.
Tell us which decisions you want to bring under review, and which systems hold the data.
Write to us with the form, or by email, and we will reply by email.
We will reply by email.
Tell us about the decision you want to bring under review.
Email us →