Sovereign AI Agents: Private Models, Private Compute, Human Control

Data sovereignty is no longer enough. In the age of LLMs and AI agents, organizations also need intelligence sovereignty: control over the models, compute, context, tools, memory, policies, and decisions that act on their data.
This hands-on workshop shows how to build a small AI Sovereignty Runtime using Free and Open Source principles. Participants will work with a concrete architecture based on three layers: private models, private compute, and private agents. The runtime will demonstrate how sensitive inputs such as logs, vulnerability reports, internal documents, prompts, and operational requests can be processed through a local-first AI control layer before reaching an LLM or agent workflow.
Participants will build and reason through a practical flow: an input enters the runtime, privacy checks detect secrets and PII, policy-as-code defines what the agent is allowed to do, a local/open model performs controlled analysis, tool access is restricted, decisions are logged for auditability, and human approval is required before any sensitive action is accepted.
The workshop connects FOSS engineering with Europe’s digital sovereignty direction and AI governance needs. It translates abstract ideas such as trustworthy AI, privacy-by-design, human oversight, data governance, and responsible automation into concrete runtime controls.
This is a vendor-neutral workshop, not a product demo. Participants will leave with a practical blueprint for building AI systems where the organization keeps control over the intelligence layer: where the model runs, what data it can see, what agents can do, what policies are enforced, what gets logged, and where human approval is required.