Sovereignty first
Infrastructure and data boundaries should be explicit design inputs, not late deployment options.
Product / What I am building
A control plane for coding agents.
A control plane for running Codex, Claude Code, Pi and OpenCode on public cloud or self-hosted infrastructure in Europe, with explicit access, execution and data boundaries.
MVP in development
The product is currently being built. Architecture and scope may evolve as the MVP takes shape.
The infrastructure question
Autonomy still needs boundaries.
- Who controls each agent?
- Where can it execute?
- Where does its data go?
- Which boundary may it cross?
- How is its behavior observed?
Conceptual architecture
Central control. Distributed execution. Explicit boundaries.
The control plane is being developed to manage policy, identity and execution for coding agents on public cloud or self-hosted infrastructure in Europe, including private cloud and on-premises environments.
A central control-plane layer connects to independent coding agents. Each agent executes inside a separately bounded public-cloud, private-cloud or on-premises environment in Europe.
Design principles
The ideas shaping the direction.
Control with context
Identity and policy should be close enough to execution to explain what is allowed and why.
Observable autonomy
Agent operation should remain inspectable without pretending that every action is deterministic.
Environment choice
Choose public cloud or self-hosted infrastructure in Europe, including private cloud and on-premises environments.
A product grounded in infrastructure work.
The questions come from years spent making physical systems, cloud platforms and production infrastructure understandable and operable. The product applies those same disciplines as agents become infrastructure actors.
Working on the same control problem?
Share your constraints and technical perspectives to help shape the MVP as it develops.