Glossary

Context layer

A context layer is the governed layer between an organization's systems and its AI agents: it unifies data into one permissioned source of context, grounds agent reasoning in current information, and logs what every agent saw and did.


Adjacent terms name parts of the job: RAG is a retrieval technique the layer operates; MCP standardizes the connection to tools and data; a vector database is one store inside it. The context layer is the whole, run as governed infrastructure for every agent rather than rebuilt per project.

It matters most in regulated industries, where every consequential agent action must be grounded in permitted, current data and reconstructable after the fact — the audit trail regulators and model-risk teams expect.

Related terms

Learn more


Sources

  1. Anthropic — Effective context engineering for AI agents (2025)