Glossary

The language of governed context

The terms behind putting AI agents into production with trust, governance and audit.

Agent observability (AgentOps): Measuring agents’ value, quality and risk continuously, closing the loop with governance and audit.
AI agent governance: Explicit, auditable controls over what each agent can use, who accesses what and how to prove what was consulted.
AI agent in a team: An agent that operates as a persistent member of a channel, squad or workflow, rather than answering one user in an isolated chat.
Approved source: Material that went through an approval queue (draft → official) with a defined owner and version, and only then can ground an answer.
Context engineering: The discipline of preparing the context layer (sources, scope, permissions, version, evidence) before plugging in any agent.
Context Quality Score: Composite index that measures whether a base is fit to feed agents: coverage, freshness, consistency, traceability, permissions/scope and gaps.
ContextOps: The continuous operation of context for agents: prepare, govern, evaluate, diagnose and improve on a recurring basis, not as a one-off project.
Enterprise MCP: The Model Context Protocol applied with governance: scope, approved sources, permissions and trail, not just the connection protocol.
Evidence Log: Auditable per-interaction record: question, context used, sources, version, scope, permission, outcome and a traceId tying it all together.
Evidence trail: The per-interaction trail that lets you reconstruct an answer’s source, version, scope and permission, the basis of auditability.
Governed context: Corporate knowledge prepared with approved source, version, scope, permission and trail, ready for an agent to consume with confidence.
Governed RAG: Retrieval-augmented generation with control: approved source, active version, scope, permission and traceability. Unlike dumping documents into a chatbot.
No source, no answer: Product principle: with no approved source, the agent refuses honestly (insufficient_context) instead of fabricating.
Scope per collection: Control that defines which sources each agent can use, applied at serving time, to prevent overly broad access.
Shadow AI: Agents and automations created by teams with no inventory, owner or source criteria. What is not in the inventory cannot be audited.
Shared context: The set of sources, rules and permissions a single agent uses to answer many people in a channel, squad or workflow.