203 - DataHub Search: describing the domain

If any term in this lesson feels unfamiliar, the 100 series is one click back. The 200 series assumes that mental model.

DataHub is an open-source metadata platform that Plexara uses as the catalog backend. Its job is to hold everything the agent needs to know about your data before it touches any data: what datasets exist, what columns they contain, who owns them, what they mean in business language, which dashboards consume them, which ones have been deprecated, and which curated queries are appropriate against them.

Plexara exposes DataHub to the agent through a toolkit of related tools. Most of them are read operations a power user cares about on an ordinary day; a smaller number are write operations that administrators and knowledge curators use to keep the catalog current. Both groups live under the same toolkit, but persona-based governance decides which ones a given caller can see.

The first tool in the toolkit is also the one used most often. A datahub_search call with a natural-language query returns ranked catalog hits, each fully enriched with the metadata the agent needs to decide what to do next. The enrichment is automatic; the agent does not make separate calls to fetch owners, tags, or glossary terms.

The operating manual that platform_info loads tells the agent to call datahub_search (and usually datahub_get_queries) before writing any query against the data. This is not a style preference. It is how the agent avoids inventing schemas, misreading column names, or picking the wrong definition of a metric.

Even with perfect catalog metadata, exploratory sessions benefit from a short domain warm-up at the start. Giving the agent a moment to describe the data estate in its own words activates the relevant slice of its training-time world knowledge and surfaces any obvious gaps before you ask a question that depends on them.

The enrichment Plexara attaches to every tool result is thorough, but it covers only what has been documented. Obvious-to-humans context (a retailer sells physical goods through stores, a bank charges fees across customer accounts, a SaaS vendor tracks monthly recurring revenue) is not in the catalog unless somebody wrote it down. A short warm-up closes that gap.

The most useful field in a catalog record is almost always the glossary term bindings. Questions phrased in business vocabulary ("show me active customers by region") reach a technical schema that almost certainly does not have a column called "active customers." The glossary term is how the agent resolves the mismatch. It is also how your organization's specific definition of "active" gets applied, rather than the one the model inferred from general training data.

Not every session needs a domain warm-up. Narrow, specific questions with clear entity references ("daily revenue by region for 2025") carry enough context in the question itself that datahub_search plus the enrichment fills in the rest. The warm-up pays off most for new users, cross-domain questions, and anything where you are not already sure which dataset you want.

With the right dataset identified and a curated query template in hand, the agent is ready to run the query. The next lesson covers how Plexara reaches data through Trino and why picking the right query shape matters so much for performance.

Six terms cover the vocabulary you will see across DataHub-related tool results, catalog documentation, and every conversation about the discover step.