Let your people and their AI build freely. On data that is right.
Analysts and data scientists increasingly work with AI assistants that produce in minutes what used to take weeks: an analysis, a model, a dashboard. That is a gain, as long as the figures underneath stand firm. Yres is the foundation that does not move: one controlled layer that fetches your data, keeps its history and makes every step traceable. Above it, things may move fast. Below it, things are boring, on purpose.
Last updated:
What changes when your team starts building with AI
An analyst working with an AI assistant produces in an afternoon what used to take a sprint. That is the attractive side. The other side: the assistant builds on whatever it finds. Point it straight at the ERP and its work breaks as soon as the vendor changes an API, and it only ever sees today's state. Ask two analysts for the same analysis and you get two sources, two definitions and two answers.
So the problem is not the AI, and not the speed. It is the layer underneath. As long as that layer is not fixed, AI multiplies the noise instead of the insight.
- Source systems change; an analysis built directly on them changes unnoticed
- Sources only show the current state; what was there last month is gone
- Every analyst picks their own definitions, and nobody sees that they differ
- A figure in a report cannot be traced back to where it came from
The foundation: one layer that does not move
Yres places one fixed layer between your source systems and everything built on top. That layer fetches the same data the same way every night, keeps every version of every row, and records where each figure came from. Analysts, dashboards and AI assistants all read from that one layer, in your own Azure environment.
The result is that experimenting becomes safe. Whoever builds on it cannot break anything in the source of truth. And what an AI assistant calculates today can still be repeated and checked tomorrow, because the data underneath has not shifted.
| What your organisation wants | What Yres puts underneath |
|---|---|
| Speed without risk: experimenting does not touch the reporting | A fixed load layer with per-table rollback; experiments read, they do not write into the source of truth |
| One version of the figures: AI answers and the monthly report come from the same source | One data warehouse with full history (SCD2) and lineage down to object level |
| Explainable and in your own hands: show where a number came from | Everything in your own Azure tenant, ISO 27001, every change recorded, including what someone does directly in the database |
| Look back: what did it look like on 1 January? | Every version of every row is kept; an analysis on the state of that day remains possible |
What this means for whoever builds on it
For a data scientist the foundation is a clean, stable dataset with history, in ordinary SQL tables and views, without first having to understand three APIs. They build their model on the layer, not on the source. If they want their own tables and views next to those of Yres, they can, in a schema of their own that travels to production through the change process.
For a BI team it means dashboards and ad-hoc analyses come from the same source, and a board member's question about a deviating figure can be answered: this is the source, this is the definition, this was the state on that date.
For an executive it is the guarantee that the speed of AI does not come at the expense of the reliability of the figures decisions rest on. What the auditor or the regulator asks for can be shown.
On the roadmap: an MCP server for AI assistants
Yres will get its own MCP server in October 2027. MCP is the open standard through which AI assistants such as Claude and ChatGPT talk directly to a system. An assistant can then query the Yres layer itself: which tables exist, what they mean, how current they are, and the data itself, within the permissions the organisation grants.
The idea stays the same as today: the assistant builds on the foundation, not directly on the sources. Only then no person needs to sit in between to translate the question into the right table. Until then, any AI assistant simply works through SQL on the layer, like any other tool.
Start today
You do not need an AI strategy to start with this. You need a layer that is right. Connect your sources, let the history build up, and give your analysts and their assistants access to the layer instead of to the sources. From that moment on, every analysis, every model and every experiment grows on the same foundation, and whatever changes in the tools above, the figures underneath stay put.
Frequently asked questions
Can an AI assistant work with Yres today?
Yes, through SQL. The Yres layer consists of ordinary tables and views in an Azure SQL database in your own tenant. Any tool or assistant that speaks SQL can read from it today, within the permissions you grant. The MCP server planned for October 2027 makes that contact more direct, but is not a prerequisite.
Isn't this just a data warehouse?
Yes, at its core it is. What is new is not the layer but what happens on top of it: AI assistants building at high speed make the demands on that layer stricter. A data warehouse that sometimes runs a day behind or keeps no history used to be a nuisance; with AI on top it becomes a source of errors that spread quickly.
What if an analyst builds something that is wrong?
Then the damage is limited to that one piece of work. The Yres layer is read-only for whoever builds on it; nobody can accidentally alter the source of truth with it. And because the data underneath is fixed and has history, the error can be traced and repeated, which is the first step to finding it.
Where does the data the AI reads live?
In your own Azure tenant, in the region you choose. Yres runs as Azure Data Factory pipelines and an Azure SQL database in your subscription. Yres itself only holds the settings of the web app, never customer data, credentials or secrets.
Does Yres generate the data model for my data scientists?
No, deliberately not. Yres delivers the layer with full history; the model on top you build yourself, in the method that suits your team. Modelling depends heavily on what your reporting tools and your people need, and that changes. The foundation does not.
