Stack layers

The agent-ready vertical stack: from contracts to evidence.

• The discovery-to-governance control plane — The end-to-end lifecycle that turns a tool-calling model into governed agent execution.
• Contracts and interfaces — Authoritative, machine-readable descriptions of what a capability is and how to call it.
• Discovery and current-context retrieval — Finding that a capability exists, and selecting a small candidate set at inference time.
• Understanding and grounding — Schema grounding vs semantic grounding: matching types vs matching intent.
• Identity and authorisation — Establishing who or what is acting, for whom, for what purpose, with which limits.
• Execution and orchestration — Durable, stateful multi-step execution owned by the runtime, not the prompt.
• Policy enforcement — Decoupling policy decisions from enforcement points across the stack.
• Observability and provenance — The evidence layer: traces, metrics, logs and an interoperable provenance graph.
• The federation layer (the missing plane) — The collaborative-learning plane that lets institutions improve a shared model without centralising data.