Data sources

Connect your warehouse so Adverity Atlas has data to query. It discovers your schema, profiles your fields, and uses that context to write better queries.

You can connect several data sources and decide which workspaces each one is bound to.

Connect a warehouse

Adverity Atlas queries your warehouse in place. It only issues read queries against your tables, it does not create, modify, or drop them, so a read-only role is all it needs.

Connections are added from Settings > Data Governance > Data Sources. Under Add Connection, pick a platform and the Add a data source wizard walks you through five steps:

1. Source: Pick the platform

2. Auth method: Choose how to authenticate

3. Authenticate: Provide configuration and credentials

4. Test: Prove the connection works

5. Review & create: Name it, schedule it, and create it

On the Authenticate step you do not have to hold the credentials yourself. Under Or ask someone else, Share a one-time link sends the credential form to whoever does, and the wizard moves on to Test when they submit.

The Test step tests the connection as soon as you reach it, and Continue unlocks only once the test passes. Click Test again after fixing a failure, or Back to enter the credentials again.

On Review & create you also pick a Mode, which decides how automated operations treat the source:

Mode	What it means
Active	Sync runs on schedule and agents can query this source
Development	Sync runs, but agents cannot query it yet, useful while iterating
Disabled	No automated operations. The row stays for history

Adverity Atlas verifies the connection as soon as it is created, and starts the first sync when verification succeeds and the mode allows it.

Each connector has its own page with the exact setup steps, required provider-side roles, and every connection field:

Connector	Authentication methods
Google BigQuery	Service Account JSON, Workload Identity Federation, Sign in with Google
Snowflake	Username & Password, Key Pair, Snowflake OAuth
PostgreSQL	Username and password
Amazon Athena	IAM Role (default credential chain), Access Keys, Assume Role (STS)
Databricks	Access Token (PAT), OAuth M2M, Azure Service Principal
Amazon Redshift	Username & Password, IAM Role (default credential chain), IAM Access Keys, IAM Assume Role (STS)
Trino / Starburst	Username & Password, JWT Token, OAuth 2.0 (client credentials)

Note

Amazon Athena is the one exception to read-only. AWS writes every query result to the S3 staging directory you configure, so that identity also needs write access to that bucket. Your tables are still only read.

Schema discovery and profiling

On first connection, Adverity Atlas scans your warehouse and profiles every field. This is what powers glossary classification and query accuracy.

• Schema discovery: Tables, views, columns, data types, and relationships, detected on first sync and on every refresh

• Field profiles: Value distribution, null rate, cardinality, min/max, percentiles, and sample values

• Include and exclude patterns: Filter which datasets or schemas are synced, useful for large warehouses

• Refreshing: Syncs run on a schedule you set; trigger a manual refresh after a significant schema change

Discovery and profiling are what glossary classification works from. Once a source has been profiled, Classify matches its columns to glossary concepts, and Extract Knowledge proposes knowledge items from what classification found. Both run from the source in Data governance, in that order.

Browsing and curating your schema

What Adverity Atlas discovered is browsable from Data > Sources in your workspace. The page lists each bound source with its Type, Status (Healthy, Degraded, Unhealthy, or Unknown), and Last sync. Open a source to browse it.

Inside a source, choose a schema from the collection picker, and its tables and views are listed with Name, Type (Table, View, or MV for a materialized view), Columns, and Rows. Two selects narrow the list: All assets or Tables only, and Visible & hidden, Visible only, or Hidden only.

Click an asset’s row to expand it. Selecting text in the row, for example to copy a table name, does not expand or collapse it. The expanded row lists the asset’s columns with Name, Type, Role, Concept (the glossary concept the column was classified to), Platform, Nullable, PK, Null Rate, and Distinct.

On a table with more than ten columns, two controls sit above the list: Filter columns… narrows it by name, and Excluded only limits it to the columns Adverity Atlas has left out of classification. Both filter the whole table, not just the rows scrolled into view.

Click a column to open its profile. The header names the collection and table the column belongs to, above the column name. The profile shows Null Rate, Distinct, Total Rows, Value Distribution, Statistics (min, max, mean, median, standard deviation, and percentiles), Top Values, and Sample Values. Where a column has never been profiled, Run Profiling starts one; Refresh re-profiles a column that has.

Reading the Role column

The Role column says where each column stands, so a table that is only partly classified reads as expected rather than as broken:

What you see	What it means
A role badge (Metric, Dimension, Identifier, Timestamp, Text / Label, or Unknown)	The column has been classified
Excluded	Left out on purpose. Hover it for the reason: every binding hides this table, or profiling found every value NULL
A dash	Still queued for classification

A column that was genuinely classified keeps showing its role even if its table is hidden later, or the column later goes empty. Only a column that was never classified reads as Excluded, which is why Excluded only is a reliable way to see what classification has skipped.

Note

Classification never covers the Adverity system columns, those whose names start with _SYS_ (such as _SYS_ID) in tables named ALL_FACTS, FACTS, or INVALID_FACTS. Both names are matched regardless of case. These columns keep showing a dash and do not link to a glossary concept. A _SYS_ column in any other table is classified as usual.

Relationships

Beneath an asset’s columns, Relationships lists the joins Adverity Atlas detected for it: a Constraint name, the Source and Target (each naming the asset and the fields joined), and the Cardinality as 1:1, 1:N, N:1, or N:M. Click Confirm on one you have checked and it is marked Confirmed, recording that the join is right rather than merely inferred. An asset with none says “No relationships”.

Hiding what nobody needs

Not every table in a warehouse belongs in an answer. Toggle an asset’s visibility from its row, or tick several and use Hide selected and Show selected. Hide entire collection does the same for a whole schema at once, and Show entire collection reverses it.

Hiding a collection is a standing rule, not a snapshot of the tables it holds today. A table the warehouse gains later is hidden from chat as soon as the next sync picks it up, without you going back to hide the collection again. While a collection is hidden in every workspace the source is bound to, Adverity Atlas does not introspect it at all, so its new tables do not appear until you show the collection, at which point the next sync catches up on everything added meanwhile.

The flip side is that a table inside a hidden collection cannot be shown on its own. Its row carries a disabled control reading “Hidden with its collection, show the entire collection to change it”.

Hiding is not only a display choice. A hidden asset is excluded from chat, and once it is hidden in every workspace the source is bound to, Adverity Atlas also stops refreshing its schema and profiling its columns. Syncs get faster and cheaper in proportion to how much you have hidden, and no warehouse queries are spent on tables nobody can read.

Note

Hiding an asset in one workspace never degrades another, the skip only applies once it is hidden everywhere. Making an asset visible again schedules a fresh profile straight away and forces a full schema re-read on the next sync, so it comes back accurate rather than stale. A table that appears in your warehouse for the first time is always read once.

Columns hide the same way, from the toggle at the end of each column row. A hidden column stays out of chat while the rest of its table remains available, which is the tool for a table that is useful apart from one field nobody should be answering from.

Adverity Atlas points out the obvious candidates itself. Where a table has columns that are all-null, empty, constant, or duplicate, a banner above the column list says how many look safe to hide, and Hide all hides that set in one click.

Tip

Profile selected runs profiling for a batch of assets in one go. Reach for it after connecting a large warehouse, rather than opening columns one at a time.

Managing connections

Once a source is connected, ongoing management happens in Settings:

• Edit credentials, update authentication when keys rotate or passwords change. Clicking Update checks the new values and tests the connection before anything is saved: a malformed value is reported by name, and a connection that fails reads “Connection verification failed. Check the host and credentials.”

• Configure sync schedule, control how often the schema is re-scanned.

• Workspace binding, control which workspaces can query which data sources.

• Monitor connection health, check status and run connection diagnostics from the source panel, see below.

Connection diagnostics

When a source turns unhealthy, “it does not work” is rarely enough to act on. Diagnose probes the live connection step by step and tells you which permission is missing, not just that something failed.

Open the source under Settings > Data Governance > Data Sources and click Diagnose in its panel. Connection diagnostics opens and runs in the background, reporting “Probing the live connection. This can take a few seconds.” while it works.

Four probes run in order, each one proving a different grant:

Check	What it proves
Connect & authenticate	The credentials are accepted and the role resolves
List schemas	The role can see the database and its schemas
List tables in schema	The role has usage on the target schema
Read a table	The select grant actually works on an object in it

On Snowflake a fifth check, Warehouse usable, runs after authentication. Snowflake drops an unknown or unauthorized warehouse silently at login, which leaves the session with no compute, and this is the check that catches it.

The result opens with an overall verdict: every check passed, connected but with limited visibility, or a check failed because the connecting role is missing access. Each row below it shows its own status, how long it took in milliseconds, and a one-line detail. A failed step adds:

• Suggested fix: The statement that grants what is missing, ready to copy. Shown where Adverity Atlas can work out the specific grant from the error

• Raw driver error: The warehouse’s own error message, collapsed until you open it

Re-run repeats the diagnostic after you have applied a fix, and Back to source returns to the source panel.

A source that is already degraded or unhealthy explains itself without a diagnostic: its panel carries a Why degraded or Why unhealthy block with the recorded error, and a See troubleshooting steps link that opens the troubleshooting table on that connector’s own page.

Note

The same checklist runs on the wizard’s Test step, and you cannot continue to Review & create until every check passes. A wrong password or a missing grant is resolved during setup rather than becoming an unhealthy source afterwards.

Running a diagnostic needs the Editor role, the same as triggering a sync.

Troubleshooting

Symptom	Likely cause
Connection test passes but queries fail	The identity has metadata access but not query permissions. Check that it can both list and select.
No tables visible in a child workspace	The workspace binding does not include those schemas, or the role lacks permission for that scope.
Data appears stale	Check the sync schedule and last sync time, then trigger a manual refresh.

Files

Files are data sources too, for data that does not live in a warehouse, budget spreadsheets, targets, campaign plans. Upload CSVs and spreadsheets and query them alongside your warehouse tables.

See Files for uploading and managing them.

Permissions

Roles are cumulative, so the role named is the lowest one that can take the action.

Action	Required role
Browse connected sources, schemas, and field profiles	Viewer
Query a bound source in Chat	Analyst
Connect a data source, edit it, trigger a sync, run a diagnostic	Editor
Bind or unbind a source to a workspace, delete a source	Admin

What’s next?

• Configure the glossary, align the language across your platforms so answers use your definitions

• Ask questions about your data, start a conversation against what you just connected

• Files, upload CSVs and spreadsheets for data that is not in the warehouse

Was this page helpful? Yes No