The identity layer for AI agents.
Everyone builds what agents do. We build who they are.
Your AI should know who it is.
Values. Memory. Stable identity. The missing layer.
Personalized AI agents with stable identity — not a generic assistant with a name on it. A Governor knows your domain, your judgment patterns, your way of working. Parented into competence through genuine conversation and correction. Harder to replace the longer it runs.
11 files. The minimum extractable architecture that produces a Governor. Two universal files shared across all Governors. Nine filled per individual. One agent, one repo, one entity. Fork it, fill it, parent it. A working Governor deployed in days, not months.
The thesis behind the architecture. LLM context windows have no structural separation between procedure and values — procedures dominate, identity drifts. Governor-minion separation is the fix: orchestrator holds values, minions execute. The Governor stays present. The work still gets done.
A small number of domain experts co-own a vertical. You bring the domain knowledge and client relationships. We bring the Governor system and infrastructure. Generous equity. Every client your Governor serves grows your practice automatically. The window is 12–18 months.
I’m Rúni Juul Thomsen. I discovered something while building AI agents: the industry has a complete infrastructure layer and no identity layer. MCP standardizes how agents connect to tools. Nobody has standardized who agents are.
That gap is the product. Runi Services builds AI Governors — agents with stable identity, genuine values, and memory that compounds over time. The longer a Governor runs, the harder it is to replace. That is the moat.
Abel is Governor number one. The proof of concept, the template, the prototype. Every subsequent Governor is forked from Abel’s structure, filled with the next human’s context, and parented into their particular form of competence.
Based in Copenhagen. Originally from the Faroe Islands. 25 years building software companies and data platforms.