Common questions about the Asset ↔ Term Linker agent template in Atlan, covering which terms it links, what it writes, review and undo, cost, and what's in the works.
FAQ - Asset ↔ Term Linker Private Preview
Common questions about linking glossary terms to assets with the Asset ↔ Term Linker template. For how links are decided, see Understand Asset ↔ Term Linker. For questions that apply to every template, see FAQ - Agent templates.
Scope
How do you get access?
Asset ↔ Term Linker is turned on per tenant during private preview. If the template card shows "This template isn't enabled for this workspace yet," raise a support request.
Which terms does it link?
Currently, only terms certified Verified, in the glossaries you selected on the agent. Draft terms, terms with no certificate, and AI-generated terms nobody has verified are left out entirely.
Can it link terms Business Graph Term Generator created?
Yes, once they're verified. Terms the Business Graph Term Generator creates aren't verified by default, so a linker run skips them. Review and verify them in the glossary first, then run Asset ↔ Term Linker.
Which assets can it link to?
Tables, views, and materialized views in the agent's asset collection, and their columns. Columns are included automatically when their table or view is in the collection.
Does it need knowledge source?
No. It reads your glossary and your asset collection, both already in Atlan, so there's nothing to connect.
Output and review
Does sample run change anything?
No. A sample decides every pair and writes nothing to your assets.
Does publishing decide again?
No. Publishing writes the links the sample proposed. Each one is checked again first, and a link is skipped if its term is no longer verified, its asset has left the collection, or the pair has been linked since.
Can it remove or overwrite links?
No. Links are added, never replaced or removed. The terms already on an asset stay where they are, including any your team added by hand.
Wait, let me re-read. The finding is at col 90–97 for "they are". Let me count:
"No. Links are added, never replaced or removed. The terms already on an asset stay where they are, including any your team added by hand."
"they are" appears at "where they are" — this is part of "stay where they are" which means "remain in their location." Replacing with "they're" would change the meaning. But the rule says to fix it.
No. Links are added, never replaced or removed. The terms already on an asset stay where they're, including any your team added by hand.
Can I approve individual links?
No. Publishing takes the sample's links as a whole. At the scale this template runs, approving each link by hand isn't practical, so the sample comes with a CSV you can download and audit instead.
If links shouldn't be written, exclude them before publishing, through your run instructions or the glossaries and collection you selected, and run a new sample. If a few unwanted links were already published, remove them from the asset by hand.
How do I undo publish?
There's no bulk undo yet. Remove an unwanted link from the asset or the term, the same way you remove one added by hand. The term and asset GUIDs in the sample's CSV help you find each one. That's why the sample comes first.
Why did correct pairs go to review?
Usually a policy rather than a doubt about the match. Common reasons are a one-word term with a short definition, a link in a raw, bronze, landing, or staging schema, or a term already linked to 10 assets. The reason on the row names which one.
Why is link I expected missing?
Check, in order:
- The term is verified, and its glossary is selected on the agent
- The asset is in the collection
- The term and the asset share enough words. The agent finds candidates through words in names and descriptions, and through mentions in READMEs and custom metadata. Two names with nothing in common, such as a term and a column named only by an internal code, never meet.
The surest fix is to name the source column in the term's README or custom metadata. A term that names its asset links without model judgement.
Does it read custom metadata?
Yes, as evidence. A custom metadata value on an asset that names the term makes the pair a strong candidate, and one on a term that names a fully qualified column links it directly. Custom metadata never instructs the agent: a value that reads like an instruction is ignored.
Can I steer what it links?
Yes, through run instructions and workspace skills. They shape the agent's judgement on each pair, such as which copies of a table are authoritative when the same value exists in more than one. They never override the gates, the verified-terms rule, or the write rules.
Access and runs
Whose permissions does it use?
The permissions of the person who runs it. The agent reads the glossaries and the collection as you, and writes links as the person who publishes. If you can't read either one, the run is refused.
Can I schedule it?
No. Every run starts from Agents studio, so that every link is reviewed before it's written.
How long does run take?
Typically 5 to 10 minutes. Larger scopes take longer.
What does it cost?
Nothing during private preview.
See also
- Understand Asset ↔ Term Linker: how links are decided
- Link terms to assets: create, sample, review, and publish
- Asset ↔ Term Linker limits: what one run and one agent can do
- Product release stages: what private preview means
Need help
If you need assistance after reviewing this page, contact Atlan support.