Atlan Release Notes Product Updates logo
Back to Homepage Subscribe to Updates

Product Updates

See the latest new features, improvements, and product updates

Labels

  • All Posts
  • Assets
  • Glossary
  • Insights
  • Workflows
  • Governance
  • Admin & Integrations
  • Reporting Center
  • Developers
  • Fix
  • Improvement
  • New
  • AI Development Lifecycle
  • Context Agents
  • Context Engineering

Jump to Month

  • September 2026
  • August 2026
  • July 2026
  • June 2026
  • May 2026
  • April 2026
  • March 2026
  • February 2026
  • January 2026
  • December 2025
  • November 2025
  • October 2025
  • September 2025
  • August 2025
  • July 2025
  • June 2025
  • May 2025
  • April 2025
  • March 2025
  • February 2025
  • January 2025
  • December 2024
  • November 2024
  • October 2024
  • September 2024
  • August 2024
  • July 2024
  • June 2024
  • May 2024
  • April 2024
  • March 2024
  • February 2024
  • January 2024
  • December 2023
  • November 2023
  • October 2023
  • September 2023
  • August 2023
  • July 2023
  • June 2023
  • May 2023
  • April 2023
  • March 2023
  • February 2023
  • January 2023
  • December 2022
  • November 2022
  • October 2022
  • September 2022
  • August 2022
  • June 2022
  • May 2022
New Context Agents
today

Context Agents now read unstructured sources, from Knowledge Files to Confluence, Git, and beyond

🎉 What's new

Your team already knows things about your data that can't be inferred from a column name: that this table refreshes at 06:00 UTC, excludes test orders, or that a null here means "not yet reconciled," not zero.

You may already have that context in Atlan as a Knowledge File.

But until now, the Context Agents that drive enrichment across so much of your data estate couldn't read it. They enriched purely from schema and lineage, with no way to point them at the document that already had the answer.

That changes now, with two new offerings.

First, link a Knowledge File to an asset (GA), and the Description and README context agents read it the next time they enrich that asset, grounding what they write in what your team already documented instead of inferring it. SOPs, best practices, data contracts, data dictionaries, runbooks, migration notes: all of this unstructured and semi-structured information can now shape how Atlan describes your data.

✨ Let's dig deeper

  • Link from two places on the asset profile. Use the Knowledge files section in the details panel, or the + in the resources row under the Readme. Either opens the same picker: search, select one or more files, and click Link files.
  • Scoped by design. An agent reads only the files linked to the asset it is enriching, never all of them. Linking is how you decide which document applies where.
  • Written, not copied. Agents take the facts a document establishes and express them for the asset. You get the same facts in different words, not a trimmed passage.
  • Nothing gets overwritten. Enrichment fills gaps. Descriptions and READMEs you already have stay exactly as they are.
  • The document has to be a Knowledge File in Atlan. If yours isn't uploaded yet, create a Knowledge Folder and upload it first.

Second, and launching in Private Preview: Agent Templates

Everything above works in one direction: an agent reads a document you linked and writes onto an asset you already have.

Agent Templates run the other way. Instead of enriching what's in Atlan from schema and lineage, an agent reaches the systems where your documentation already lives, reads what's written there, and creates governed context that wasn't in Atlan at all - citing its source every time.

Each template is built for one outcome. You pick the outcome you want, connect the source, say what you're after, preview, then run. The first one is live:

Business Graph Term Generator turns your docs into business graph terms and categories, enriches terms that have no definition, and links them to the assets they describe. Every term carries a citation back to the document it came from.

  • You supply four things, and none of them is the prompt. The source to read, the scope inside it, run instructions saying which terms you want, and optionally your own skills for house rules a template can't know. Atlan packages the rest: the skills that make a run reliable, the quality gates it has to pass, and the write path into Atlan.
  • Sample before you publish. A sample run reads a small part of your sources and proposes a handful of terms in a minute or two. It publishes nothing. Read it, adjust your instructions, sample again, then run for real.
  • Your Knowledge Files are a source here too. The documents you're linking to assets today can also be what a template reads from, with nothing to connect and no credentials to manage.
  • Read-only by construction. For GitHub and Confluence, an admin connects the source once in Settings with a read-only credential you provide. That credential is the access boundary: the agent reads exactly what the token can read, can't write back to the source, and never asks an end user to sign in.
  • More outcomes are coming. Selecting the bell on a template card records your interest, and that signal decides which one gets built next.

Agent Templates are available in the New UI only, enabled per tenant, and open to workspace admins and governance admins. Runs are free for the duration of the preview. If you don't see a Create agent button in Agents Studio, your tenant isn't switched on yet - talk to your Atlan contact!

New to this? Start with Understand agent templates.