Kernaro AI Hub vs Manual Data Export: Getting Current Architecture Data to the People Who Use It

Part of an architect’s job is getting architecture facts to the people who decide: portfolio owners, project sponsors, review boards, and now the AI assistants those people ask. In most teams that still means exporting from Sparx EA. Every Excel sheet, CSV, generated report and slide is accurate when it is made and out of date soon after. For anything used more than once, that drift is the problem.

This piece compares the manual export with a live connection through Kernaro AI Hub, a Sparx Systems product. It covers what changes for the architect, what the new way asks of them, where a dated export is still the right answer, and what to do if you do not have the Hub.

How architects get data out today

Most teams lean on some mix of four mechanisms.

CSV and Excel export. The built-in exports produce files of elements, attributes and tagged values. The files land in email, SharePoint or a wiki, where they sit and age.

Report generation. The built-in report generator turns the model into RTF or PDF documents. That suits formal deliverables such as architecture definition documents and transition plans. It suits anything that needs to stay current poorly.

Slides. Diagram images go into decks for governance meetings. The deck is correct on the day it was assembled.

Custom scripts. More mature teams write PowerShell or Python against the repository database or the automation API and push data elsewhere. The scripts work until the schema shifts, then they break and need a maintainer.

All four share one weakness. Once an export is generated, it starts to diverge from the repository. Change an element’s lifecycle status the day after an export and everyone downstream works from the wrong picture until the next cycle runs.

What stale data costs

Architecture data goes out of date faster than most teams admit. A few patterns recur.

Lifecycle changes. A vendor publishes an end-of-life date and the team updates the repository. The portfolio spreadsheet that went out last week still reads “Active”, and next quarter’s roadmap is built on the old status.

Ownership changes. A system changes hands in the repository. The diagram handed to the new divisional head still names the previous owner, so questions go to the wrong person.

Decision changes. An Architecture Review Board moves a decision from “Proposed” to “Rejected”. The decision register exported for the sponsor last month still says “Proposed”, and the project proceeds on a false premise.

Capability changes. A new application joins the portfolio and its coverage is modeled. The capability gap analysis in the business case does not reflect it, because it was built from last month’s export.

Decisions made on stale data create remediation work, and that work uses up the capacity the team needed for the next thing.

The AI-augmented way: people and AI ask the live model

With a live connection, nobody exports. A stakeholder asks a question, or a Power BI report refreshes, or an AI assistant looks something up, and the answer comes from the repository as it stands when the question is asked. The architect stops running the export cycle and spends that time on the model the answers come from.

What the new way asks of the architect

Keeping the modeling language meaningful. The Hub reads the repository through the structure your MDG Technology defines, so the quality of every live answer depends on it. When the metamodel changes materially, the MDG needs governance. The AI-augmented way to keep it tight is for the architect to design the governance intent, the stereotypes, required tagged values and rules, and to have AI apply and check it across the model.

Deciding who sees what. Access and authentication become a one-time setup on the Hub rather than a per-report concern. Someone still has to decide which packages each audience can reach.

Knowing when a dated copy is the right answer. Some uses need a frozen, signed-off artifact, and the architect has to recognize them (see below).

Report definitions in Power BI still need updating when reporting requirements change, which is ordinary BI maintenance. The upkeep that disappears is the export plumbing: the scripts, templates, distribution lists, schedules and the question of which copy a project manager saw.

Tools that help

Kernaro AI Hub. A Sparx Systems web product, launched in January 2026 and bring-your-own LLM: your organization supplies the keys and pays the provider for usage. It sits between the repository and the people and tools that use it. AI assistants reach it through its read-only MCP server, and other systems through its read-only GraphQL API. Because the GraphQL layer follows your MDG, the semantic layer your modeling language sets up in the repository carries through to the query.

Kernaro Connector for Power BI. An extension to Kernaro AI Hub, and our preferred route to Power BI when your organization has the Hub.

A reporting database, if you do not have the Hub. We can build a reporting database on top of your Sparx EA repository. If you host your repository yourself, we build the reporting views in the same database and you read from them live. If you cannot reach the database, or it is hosted elsewhere, and you need the reporting database in something like Snowflake or Microsoft Fabric, we help you set it up so you refresh it from your EA model on the schedule you define. Power BI then reads the model that AI Power Tools for Sparx EA, free from NovoCircle, generates from the reporting database. The reporting and Power BI page sets out how we can help, and the AI tools for Sparx EA comparison shows where each option stops.

For Microsoft 365 Copilot, we can help you get there through a published Power BI semantic model, or with the Hub through its read-only API via Copilot Studio or a Microsoft 365 connector.

Where manual export still earns its place

A live connection is the wrong tool in four situations.

Regulated reporting with an audit trail. Some frameworks, such as ISO 27001 audits and financial-services regulatory reporting, require a point-in-time snapshot that can be dated, signed off and stored as evidence. A live dashboard cannot do that. A generated report can.

Offline delivery. A client with no access to your BI environment can still read an exported document or diagram. A live dashboard needs connectivity.

One-off requests. “Can you send me a list of every application in the Finance domain?” is a bounded, one-time ask, and a quick CSV answers it.

Systems that only take files. If a downstream system can only ingest files, export is the only mechanism available.

A simple rule covers the rest. If the data will be used more than once, or its currency matters each time it is used, connect it live. If it is used once, for a bounded purpose, and a dated copy is fine, export it.

The case for going live

Better decisions. A single decision made on a stale picture, such as investment aimed at a capability already covered or a project running on a superseded architecture decision, can cost more than putting a live connection in place.

Architect time back. Time spent generating, formatting and distributing exports is time not spent on architecture.

AI that answers from the model. Serving the repository over MCP lets an assistant answer questions from your live architecture data. The Hub does that on the server side, read-only, for everyone who needs answers. AI Power Tools does it locally, with read and write, for an architect directing Claude or GitHub Copilot on their own machine. The Ecosystem Explorer shows how the pieces fit.

Frequently asked questions

For the choice of BI platform on top of the live connection, see Power BI vs Tableau for EA analytics. To shape the rollout itself, Configure the Solution is where we select the combination, connect it to your Sparx EA and prove it on your data.