How data becomes available

Data goes through four stages before another module can use it. Most confusion about Data Hub comes from expecting a later stage to work when an earlier one has not happened.

1. A connection

A connection holds the address and credentials for a database, file, or API. Creating one is a platform-level action, and linking it to a service is what makes it appear in that service’s Data Hub.

At this stage nothing is readable yet. The connection either reports Online or it does not.

2. Tables, hidden by default

Once a connection is online, Data Hub lists the tables it can see. Every one starts hidden.

Hidden means the service cannot use it. You enable the tables the service should have, individually or with Show All. This is the security boundary: a connection to a production database does not hand the whole database to everyone who can open the service.

It is also the stage people skip. A perfectly healthy connection with nothing enabled looks, from every other module, exactly like no connection at all.

3. Managed datasets

Enabling a table registers it as a managed dataset for the service. You do not create these by hand; enabling is what creates them.

The managed dataset is the thing other modules actually bind to. When a chart in Dashboards lists a source, it is listing managed datasets and views, not raw tables.

4. Views, optionally

A view is a saved query over one or more datasets — filtered, sorted, joined, or reduced to the columns that matter.

Views are optional. A dataset works on its own. But a view is usually the better thing to hand downstream, because the work of getting the shape right happens once, in one place, rather than in every chart that uses it.

Views carry a version number and show as LIVE once saved.

Where it stops

Each stage has a distinct symptom when it fails, which makes diagnosis quick if you know the order.

If the connection shows Offline, stage 1 — the credentials, host, or file path are wrong.

If the connection is online but has no tables, stage 1 succeeded against an empty or inaccessible schema.

If tables are listed but another module sees nothing, stage 2 — nothing has been enabled.

If a dataset exists but the numbers are the wrong shape, stage 4 — you want a view.

Data is read, not copied

For SQL sources, Data Hub queries the source when something asks for rows. It does not keep a copy, so what you see is what the database currently holds, subject to the cache settings on that dataset.

API-backed sources — Airtable, Coda, Notion — work differently. Their data is synced into landed tables, so there is a copy, and it is as fresh as the last sync.

That difference matters when someone asks “is this live?”. For a Postgres connection, yes. For Notion, it is as live as the sync.

Connect a source covers stage 1, Choose which tables to expose covers stages 2 and 3, and Build a view covers stage 4.