Run It Anywhere, Take It With You: SovereignTech for the Semantic Layer

One of Gartner's Top Strategic Technology Trends for 2027 is Sovereign Tech, or SovTech: hardware, software, databases and AI models "technically and legally architected to offer realistic exit strategies from critical providers." Gartner predicts that by 2030, over 75% of non-U.S.-based organizations will adopt digital sovereignty strategies. It is the second trend on that list that lands on the semantic layer; we covered AI Context Platforms two weeks ago.

Most sovereignty conversations stop at data residency. Where is the data stored, which jurisdiction governs it, who can compel access to it. Those questions matter, and they are the easier half.

The harder half is meaning. If you left your warehouse vendor tomorrow, could you take your definition of revenue with you? And if an AI agent can query that definition, where does the system serving it run, and who else can reach it?

For the semantic layer, a realistic exit strategy has two parts. The definitions have to be portable. And the platform that holds them has to run where you decide.

Take it with you

Lock-in at the semantic layer is quieter than lock-in at the storage layer. Moving data is a project with a budget and a plan. Moving meaning usually isn't in the plan at all, because the meaning was never written down in one place.

Metric definitions end up in a warehouse's native semantic views, a BI tool's modeling layer, YAML files in a transformation project, and the prompt instructions for an agent. Each of those formats belongs to one vendor. When you change platforms, the definitions don't migrate. Someone rebuilds them, and then someone else has to prove the rebuilt versions return the same numbers as the old ones.

That second step is where exit strategies stall. A revenue figure that comes out slightly different after a migration is a finance problem and an audit problem. Now that agents query definitions directly, the wrong number reaches everyone who asks an agent, not only the people who read one dashboard.

Portability starts with the format.

Coginiti 26.10 added import and export for Apache Ossie, an open, vendor-neutral format for exchanging semantic models (incubating, formerly Open Semantic Interchange). Ossie models come into Coginiti catalog projects as native semantic entities, and Coginiti semantic models go out as Ossie. Definitions can move in either direction.

The same release added Databricks, Oracle, Db2 and Netezza as semantic platforms. Coginiti connects to 21+ data platforms, so one governed definition can run against more than one engine instead of being tied to wherever the data sits today.

An open format doesn't finish the job. Ossie is still incubating, and a definition that moves to a new engine still has to be checked against the numbers it produced before. That check is easier when definitions carry their own tests, which is why we keep arguing for testing meaning, not just rows.

Run it anywhere

The second part of the exit strategy is where the platform runs. That used to be a procurement and compliance question. Agents have made it a security question.

CrowdStrike's 2026 Threat Hunting Report found that AI agent-triggered detections now surface 2.5 times more threat leads than manually driven activity. Gartner predicts that by 2028, 25% of enterprise security breaches will trace back to AI agent abuse. And in March, Gartner predicted that by 2030, "universal semantic layers will be treated as critical infrastructure, alongside data platforms and cybersecurity."

That puts the semantic layer on the agent attack surface. It holds the logic for how revenue, margin and customer are calculated. It knows which tables join to which. Through MCP and similar interfaces, agents can now reach it directly, sometimes with write access. For many organizations, that makes it a system they want running inside a perimeter they control.

Coginiti deploys and runs anywhere on Kubernetes: public cloud, a private data center, or a fully air-gapped environment. It is the same platform in each, so moving between environments doesn't mean moving to a different product. That includes Forge, Coginiti's build-side agent. In an air-gapped install it connects to models running locally on Ollama, so the agent keeps working without a connection to an outside model provider.

26.10 added pieces that matter here:

• The semantic server can be deployed separately from the main application, so it can be scaled independently as agent query volume grows.
• Domain-level permissions let administrators scope semantic access by domain, with APIs to provision access and inspect a user's effective permissions.
• The MCP server supports OAuth for external clients, and administrators can set result limits so agent interactions stay bounded.

Coginiti works with government organizations including NAVSUP, the U.S. Army Corps of Engineers, and the Department of Homeland Security's Cybersecurity and Infrastructure Security Agency (CISA). For classified and regulated environments, the platform supports IL2–IL6 and fully air-gapped networks.

Five questions for a semantic exit strategy

1. Where do your metric definitions live today, and how many places is that?
2. Can the same definition run on more than one engine?
3. Are the definitions stored in an open format, or only in one vendor's?
4. Can the platform that serves them run inside your own perimeter: your cloud, your data center, or air-gapped?
5. Who holds the version history, and does it leave with you?

If the honest answer to any of these is "one vendor," your exit strategy covers your data but not what it means.

Gartner is right that sovereignty is moving up the enterprise agenda. For the semantic layer, it is won the same place trust is: below the waterline, with definitions in an open format, tests that travel with them, and a platform that runs wherever you decide.