Understand how a memory layer maps to Atlan's architecture, and how context repositories and the Atlan MCP server compare to vector databases and RAG.
Memory layer for AI agents
"Memory layer" is the term agent frameworks use for infrastructure that persists context across sessions. It lets a stateless model recall prior facts and organizational knowledge on its next run. Atlan doesn't ship a product with that name. Instead, two components deliver the same outcome. You build a context repository in Context Engineering Studio, and the Atlan MCP server serves that repository's context to agents at runtime.
This page maps that market term to Atlan's architecture and explains how it differs from a vector database or RAG.
Map memory layer to Atlan's architecture
Atlan delivers the outcome of a memory layer through the context layer. This is a governed tier that gives agents persistent, trustworthy memory of your data through metadata and the Model Context Protocol (MCP).
Two components work together:
- Context repository: a bounded, versioned package of governed context for one domain or use case, built in Context Engineering Studio. It holds the semantic model, business logic, verified question-and-SQL pairs, and the other assets an agent needs to answer questions correctly.
- Atlan MCP server: the interface that lets AI agents search, inspect, and update Atlan metadata from their own runtime, using that governed, real-time context.
Atlan fits governed, shared enterprise context used across many agents and runtimes. It isn't a drop-in replacement for a lightweight, per-agent personalization store such as a standalone vector database.
Memory layer vs vector database vs RAG
A memory layer isn't the same as RAG or a raw vector database.
- RAG retrieves passages from a corpus at query time.
- A vector database stores embeddings for similarity search.
- A memory layer persists and re-injects context across runs.
In Atlan's model, the governed context layer is the enterprise substrate that agents read from, with lineage, policy, ownership, and semantic meaning attached.
| Approach | Persistence | Governance | Best fit |
|---|---|---|---|
| Vector database | Per-agent or per-app | Application-level | Single-agent personalization |
| RAG pipeline | Query-time retrieval | Depends on corpus controls | Retrieval from a document corpus |
| Atlan context repository + MCP | Versioned, shared context across runs | Enterprise governance, lineage, policy, and RBAC | Multi-agent, multi-runtime enterprise deployments |