Research project / live system
Deterministic AI Kernel
Four practical agents where models propose the language and deterministic code owns policy, tools, and outcomes.
why it exists
An agent demo matters only when its business constraints survive contact with an unpredictable model.
how it works
Typed inputs, schema-checked decisions, deterministic tools, repair cycles, and explicit handoffs bound each scenario.
how it looks
A compact run record exposes steps, metrics, evidence, outcome, and human handoff without hiding behind chat prose.
ScenariosOrdering, internal knowledge, refunds, collections
Runtime proofTyped decisions, seeded tools, repair, handoff
IntegrationApplication layer over Hyper Runtime contracts