Schema Intelligence

Schema Intelligence shows how InfiniStack reads the data in your connectors: which tables are ready to query, how tables join, where data flows from, and which tables Chat should query first.

What Schema Intelligence does

Open Schema Intelligence from the Data section of the app's sidebar, between Data Models and Business Knowledge. Integrators and Admins can open it. Members and team leads don't see it. See Roles at a glance.

InfiniStack builds the schema map for you. It runs automatically when you create a connector, and again when you change a connector's credentials. Each run works through these stages, in order:

Stage	What it does
discover	Finds the tables in the connector
profile	Profiles the columns in each table
relate	Finds how tables join, within the connector
flow	Traces where each table's data comes from
describe	Writes a short purpose for each table
link	Matches the same entity across different connectors
embed	Indexes the tables so Chat can find them
draft	Drafts business objects for Business Knowledge
score	Updates the intelligence score

Each run profiles, relates and describes up to 200 tables per connector.

note

Schema Intelligence never publishes Business Knowledge on its own. The draft stage only creates drafts for you to review.

The page has a toolbar and five tabs:

• Connector: show one connector, or All connectors.
• Layer: show All layers, or only Serving, Staging, Raw or Unknown tables.
• Refresh: reload what's on the page.
• Tabs: Overview, Relationships, Lineage, Query targets and Cross-source links.

If your organization has no connectors yet, the page shows Connect your data to unlock schema intelligence. Add a connector first. See Connectors.

Intelligence score and freshness

The top of the page shows three cards:

Card	What it shows
Intelligence score	The average score of your connectors, from 0 to 100. It's 0 when you have no connectors.
Cache freshness	The average, across your connectors, of the share of each connector's tables that InfiniStack has read in the last 24 hours. Below it, Gate met (≥95%) or Gate not met (≥95%).
Connectors tracked	How many connectors are scored, and how many are meeting release gate.

Each connector's score is out of 100:

Part	Points
Tables read and up to date	30
Tables with a purpose	25
Columns profiled (target: 3 per table)	15
Relationships found	10
Lineage found	10
Tables indexed for Chat	10

A connector meets the release gate when its freshness is at least 95% and its score is at least 40. Your organization meets the gate when at least 95% of its connectors have freshness of 95% or more.

The Overview tab lists each connector under Connector coverage:

Column	What it shows
Connector	The connector's name
Score	Its intelligence score
Fresh	Its freshness, shown as a warning when the connector doesn't meet the release gate
Cached	Tables read, out of tables found
Purpose	Tables with a purpose
Edges	Relationships found, then lineage links
Bootstrap	done, the stage that's running, or —. A failed run shows its error here.

If the list says No semantic health yet., add a connector or ask an Admin to rebuild one.

Layers, relationships and lineage

Every table gets a layer, which says how ready it is to query:

Layer	What it means	How InfiniStack decides
Serving	Finished tables, ready for reports and Chat	Output of a transformation, data only flows in, or the name starts with dim_, fct_, mart_, rpt_ or agg_
Staging	Tables in the middle of a flow	Data flows in and out, or the name starts with stg_, or contains tmp_ or _staging
Raw	Data as it arrived from a source	Loaded by a pipeline, data only flows out, or the name starts with raw_
Unknown	Anything else	—

InfiniStack checks the Serving rules first, then Raw, then Staging. The Overview tab shows a legend of the layer badges.

System tables, such as information_schema or tables starting with pg_, are never offered as query targets.

The Relationships tab draws a Relationship graph of how tables join: foreign keys declared in your database, joins InfiniStack inferred, and the same entity found in two connectors. Drag a table to move it, and scroll or use the zoom in and zoom out buttons. Select Expand to fill the screen, and Collapse or Esc to go back. The graph shows up to 2000 links.

Under the graph, Link details lists every link: From table, From col, To table, To col, Kind and Confidence. Confidence is one of:

• declared: the database declares the key.
• inferred_high or inferred_low: InfiniStack inferred the join, with high or low confidence.
• manual: an Admin confirmed it.

The Lineage tab lists Data-flow lineage: where each table's data comes from, with From, To, Kind, Source and Confidence. A replicate link comes from a pipeline, and a transform link comes from a transformation. The Lineage tab always lists lineage for every connector in your organization. The Connector and Layer filters don't change it.

Query targets

Query targets are the tables Chat should query first, ranked from most to least useful. The Overview tab shows the top 5 under Top query targets. The Query targets tab shows up to 40, with Table, Layer, Rank, Entity and Purpose.

Rank starts from the table's layer:

Layer	Starting rank
Serving	1.0
Unknown	0.6
Staging	0.3
Raw	0.2

Then InfiniStack adjusts it:

• +0.05 for more than 1,000 rows.
• +0.03 for 8 or more columns.
• -0.2 for a serving table that hasn't been updated for more than 3 days.
• A raw table's rank is at most 0.25. It's left out when a serving table covers the same entity.

To see which tables fit a question, type words in Intent search, for example revenue customer. Separate words with spaces or commas. Each word that matches a table's name, purpose or tags adds 0.08, up to +0.35. Tables that closely match your words in meaning can get up to +0.15 more. Rank never goes above 1.5.

Guides

Integrators can view everything on this page. Only Admins see Rebuild schema, Clear schema, Edit and Confirm.

📄️Review & correct the schema map

Check and fix what Schema Intelligence learned about your tables.