Context Agents now link glossary terms to the right tables, views, and columns
🎉 What's new
Your glossary defines Revenue, and linking it to a column is how people and AI agents find the number. But your sales table has revenue, revenue_budget, and revenue_forecast. Link by simply name and Revenue lands on all three without clarity on the distinction.
The Asset ↔ Term Linker agent, now in Private Preview, links Revenue to the right one.
Linking by hand doesn't scale, and traditional rules or regex patterns rules miss the complexity of real world organizations.
Asset ↔ Term Linker is the second Agent Template. A decision model reads your glossaries and an asset collection, judges each pair the way a data steward would, and proposes links from verified terms to the tables, views, and columns that implement them.Â
✨ How it works
- Same shape as every template. Pick up to 10 glossaries and an asset collection, add optional run instructions or skills, preview, then publish.
- Built to avoid confident mistakes. Pairs that differ in unit, currency, period, scenario, or comparison are blocked outright. Column links win over table links, one word terms and raw or staging copies go to review
- You can make a link certain. Name the column (
db.schema.table.column) in a term's definition, README, or custom metadata and it links without model judgement. Links your team made by hand are kept and used as examples. - Safe by default. A sample decides every pair and writes nothing. Review each pair's outcome, confidence, and reason, then publish what you reviewed. Links are added as term assignments, never replaced, capped at 250 per run, and no terms, definitions, or descriptions change.
- Verified terms only. Draft and unverified terms are skipped.
Available in the New UI in Private Preview. Talk to your Atlan contact to get it switched on.
Want to dig into the details? See more here