Why this company exists
AI agents are crossing a line: from proposing actions to executing them. DecisionHypervisor exists to define and enforce the boundary between the two.
Software used to advise and humans used to act. That arrangement is ending. Agents now initiate wire transfers, modify production infrastructure, and touch clinical and industrial systems directly — at machine speed, without a human in the loop. The tooling meant to govern them still assumes someone approves everything consequential, or that logging after the fact is good enough. Neither holds anymore. DecisionHypervisor was founded to close that gap: a dedicated authority layer that decides what may execute, before anything executes.
One design decision, made deliberately
DecisionHypervisor is built on a single architectural commitment: separate the intelligence plane from the authority plane.
Models are probabilistic. They excel at proposing — drafting the transfer, writing the Terraform, routing the ticket. We let them. Proposals are cheap, reversible, and fast, and the better models get, the more of them there will be.
Execution is different. Authority must be deterministic, narrow, and revocable — granted per action, at the moment of action, by a system that cannot be talked into exceptions. That system is the kernel we build.
What exists today
Not a whitepaper. The core runtime is built, running, and tested.
A Rust decision runtime evaluates every proposed action against policy — same inputs, same decision, every time.
Proposed actions are normalized into a versioned, structured schema before evaluation. No free-text guessing at the boundary.
Every decision lands in a tamper-evident audit chain — provable after the fact, without trusting us.
Deny precedence, fail-closed enforcement, and deterministic evaluation are asserted by property-based and conformance tests — not by convention.
The execution-authority escrow model underlying the runtime is patent-pending. Details →
DecisionHypervisor is a private-preview company. We work with a small number of design partners deploying agents into financial, infrastructure, and operational pipelines — teams that would rather govern execution now than explain an incident later.
Gabriel Heinemann
Gabriel Heinemann is a systems inventor and technical founder building infrastructure for the safe, authorized deployment of AI agents.
He founded DecisionHypervisor around a central principle: capability is not authority. AI systems must prove they are authorized to act at the moment of execution—not justify their actions afterward. DecisionHypervisor turns that principle into enforceable infrastructure for governing consequential actions across software, autonomous systems, and the physical world.
“Models propose. Authority decides. DecisionHypervisor ensures that every consequential action is provably legitimate at the moment it executes—not reconstructed after the damage is done.”