Understand how context agents and agent templates read knowledge files to enrich assets and extract business rules into glossary terms
Extraction pipeline
Uploading a knowledge file makes it available to agents. It doesn't start any processing on its own.
Nothing is extracted until you run an agent, and what you get depends on which one you run. There are two routes, and they serve different purposes.
Two routes into your documents
| Route | What you do | What you get |
|---|---|---|
| Enrichment | Link a knowledge file to an asset, then run the Description or README agent over that asset | A description or README for the asset, written from the linked documents |
| Agent template | Select knowledge files as a source when you create a context agent from an agent template, then run it | Glossary terms with their business rules, each cited back to the source file |
Enrich assets from linked documents
Link a knowledge file to the assets it relates to, and the Description and README agents read it when they next enrich those assets.
Two things shape this:
- An agent reads only the files linked to the asset it's enriching. It doesn't search your whole knowledge library, so linking is how you decide which document applies where.
- Agents write in their own words, taking the facts a document establishes and expressing them for the asset rather than copying its wording.
For the procedure, see Link knowledge files to assets.
Extract business rules into glossary terms
Create a context agent from an agent template, select Knowledge Files as a source, and run it. The agent reads the documents in scope and, for each rule it identifies:
- Creates or updates a glossary term with the rule's name and a definition
- Attaches the extracted business rule text to that term
- Cites the knowledge file the rule came from
A sample run proposes terms without publishing anything, so you can check the output before committing to a full run.
For the procedure, see Create an agent from a template and Use knowledge files as knowledge source.
Business graph terms
Extracted terms appear in your Atlan glossary like any other term. The difference is their source: each term cites the knowledge file it came from, so you can always see which document a rule originated in.
Business rules are attached directly to each term.
Skills
Generating skill files from knowledge files isn't available yet. When it ships, a skill packages the rules and logic from a document in a form other agents can consume directly.
Lineage
The catalog records the processing trail on the knowledge file asset, so you can see the agent that read it and the glossary terms it produced.
See also
- Knowledge folders and files: What knowledge folders and files are as catalog assets
- Link knowledge files to assets: Feed specific documents into asset enrichment
- Use knowledge files as knowledge source: Point an agent template at your documents
- Create knowledge folder and upload files: How to upload files
- Context Agents Studio: Configure and manage context agents