Context layer
What is context layer?
A context layer is the governed tier between a data estate and AI agents that translates metadata into business meaning. It combines semantic definitions, lineage, access policy, and ownership so agents can act on data without rebuilding meaning per tool.
What distinguishes enterprise context layer from single-team?
Scale, ownership structure, and compliance requirements. A single-team build covers one domain with lightweight governance. An enterprise context layer adds federated domain ownership, a compliance overlay (access controls, versioned certification, audit logs), and compounding learning as teams evaluate and redeploy context over time.
How's context layer different from semantic layer?
A semantic layer defines metrics and dimensions for BI tools. A context layer is broader: it adds governance, lineage, ownership, and other machine-readable metadata that AI agents need to use those definitions safely at runtime. The context layer serves AI agents. The semantic layer serves BI tools.
How's context layer different from RAG?
RAG retrieves unstructured text chunks at query time and passes them into a model as prompt context. A context layer exposes structured, governed metadata about actual data assets—definitions, ownership, lineage, and trusted business meaning—and governs what answers an agent is allowed to give.
How does MCP relate to context layer?
MCP is the protocol that lets AI tools connect to Atlan and request governed context in real time. It's the transport layer, not the source of truth. The context layer is what sits behind the MCP server: the governed graph, policies, and audit trail. The Atlan MCP server exposes that governed context to MCP-compatible clients including Claude, Cursor, and ChatGPT.
Do I need to enrich context for every table?
No. Start with the assets your team actually uses. Collections, coverage views, and popularity signals help you focus on the subset that matters first.
How long does it take to build enterprise context layer?
Plan for one priority domain to reach production first. Estate-wide rollout across multiple domains is a multi-quarter program.
Does enterprise context layer help with GDPR and SOX compliance?
It supports compliance audit requirements through access controls, versioned certification, and audit logs. It doesn't replace your compliance program. It gives the program the evidence layer it needs.
What is memory layer for AI agents?
A memory layer is external infrastructure that persists context across sessions so a stateless model can recall prior interactions, learned facts, and governed organizational context at the start of a new run. In Atlan, this maps to a context repository served through the Atlan MCP server.
Can multiple AI agents share same memory?
Yes. That's one of the main reasons to use a governed shared context repository instead of isolated per-agent memory. Context repositories are designed to be portable and shared across agents and runtimes.
What can agents write back through Atlan MCP?
Documented write tools support updates to asset metadata, glossary content, tags, custom metadata, lifecycle, and data quality. Treat those documented writes as the supported write-back surface.
How can I scope memory by team or domain?
Create a separate context repository for each domain, team boundary, or agent use case where business definitions differ. Context repositories are bounded and versioned, which makes them safer to govern and easier to evolve.
Further reading
- GDPR (Regulation EU 2016/679): EU General Data Protection Regulation, 2016.
- Sarbanes-Oxley Act of 2002: United States SEC, 2002.
- Model Context Protocol specification: Anthropic, 2024.