Question 1

Do you copy our database, and where does our data actually go?

Accepted Answer

Nothing is copied. Your rows are never extracted, synced or stored on Edilitics infrastructure. When a source connects, the profiler runs server-side against your database and reads back only structure and statistics - column names, data types, null counts, distinct counts, min and max values, and top values. Your records stay in your database. Credentials are encrypted per workspace, so a key that unlocks one customer's connection cannot unlock another's, and uploaded files are virus-scanned and encrypted in isolated storage before anything reads them.

Question 2

What do we have to do to connect a database?

Accepted Answer

For a live database or warehouse, whitelist one IP address - 34.14.190.234 - in your firewall or security group, then enter your connection details. That is the only network configuration required. Uploaded files need nothing at all: CSV, Excel, JSON, Parquet, Avro, Feather, Pickle, PDF and Google Sheets are all supported by uploading directly in the connection form. Connecting tests that the source is reachable and secure; it never runs anything that changes your database.

Question 3

What is the quality grade, and how is it worked out?

Accepted Answer

Every column is scored A to F automatically the moment a source connects, on three things: completeness at 50% of the score, uniqueness at 25%, and type compliance at 25%. Completeness carries the most weight because blank data is the most common and most damaging problem - a column half empty cannot answer anything reliably no matter how well formatted the rest is. Table scores are a weighted average of their columns, with ID, relationship key and date columns weighted three times heavier because everything downstream depends on them, and internal columns starting with an underscore weighted at half.

Question 4

Why does a clean table still score badly for AI?

Accepted Answer

Because clean is not the same as understood. The AI Readiness score is half structural health, taken directly from the quality grade, and half documentation - whether each column has a description and whether a person has confirmed it. A table with perfect data quality and no descriptions scores 50% and lands at Grade D: the AI has clean data but no idea what any of it means. Descriptions written by AI alone can only contribute 10 of the 50 documentation points, so reaching Grade A, which needs 90 or above, requires your team to approve them.

Question 5

Can our analyst use a connection without seeing the password?

Accepted Answer

Yes. Share an integration with any verified member of your workspace and they can use it in their own pipelines, dashboards and chats, and read its quality scores - without the connection credentials ever being returned to them. Shared users cannot edit, delete, re-share, or trigger a refresh; only the owner and workspace admins can. Integrations can only be shared inside your own workspace, and accounts on personal email domains cannot share at all.

Question 6

What happens when someone changes a table upstream?

Accepted Answer

A daily background job re-checks live database connections that have not been profiled for seven or more days and where the underlying data has grown by more than 5% - both conditions must be true. New columns are added to the schema and picked up in the quality metrics, removed columns are detected, and scores are recalculated against the current state. Every integration card also shows what depends on it - pipelines, dashboards and chat sessions - so you can see what a change would affect before you make it.

Question 7

Does profiling read every row in our tables?

Accepted Answer

Not for large tables. Above 16,600 rows, Edilitics profiles a random sample of that size rather than the full dataset - enough for 99% statistical confidence with under 1% margin of error at population level, without loading your source database. The profiler runs server-side against your database and no data is moved to calculate the scores.

Question 8

Do our actual values ever get sent to an AI model?

Accepted Answer

No. Generating column descriptions sends structural information only: table names, column names, data types and quality statistics. Three privacy modes control what else goes with it, and this is set per connection and changeable later. Private sends structure alone. Balanced adds your organisation's industry sector, so a column called amt is read as a transaction amount in finance and a dosage amount in healthcare. Full Context additionally sends the values that appear most often, which helps when your columns hold category labels. None of the three sends a row of your data.

Question 9

How does a column description become trusted rather than guessed?

Accepted Answer

In two stages. First the AI writes one description per column from its structure and statistics. Then a person on your team reads it in the Metadata Viewer, edits it if it is wrong, and marks it validated. That second stage is what moves a column from contributing 0.2 to contributing 1.0 toward the AI Readiness score. Once validated, a description is protected from automated overrides - later refreshes leave it alone, so what your people know about your data outlasts changes to the source.

Question 10

Does a low grade stop us using the data?

Accepted Answer

No. Grades D and F raise an advisory before an AskEdi session or an Auto Generate Charts run, and the tone changes with the grade - confirmation at A or B, an informational note at C, a clear warning at D or F. It never blocks access. The point is that you know what the answer is standing on before you rely on it, not that the platform decides for you.

Question 11

Can we turn AI descriptions on after we have already connected?

Accepted Answer

Yes. If you skipped it at connection time, hover the integration card and choose Generate AI Insights, pick your AI provider and privacy mode, and confirm. Generation runs in the background and you are notified when it finishes.

Question 12

How often can we refresh the scores by hand?

Accepted Answer

Once every 24 hours per integration. A refresh is also triggered automatically when connection credentials are edited in a way that changes what is being pointed at - hostname, port, or database name. Separately, every successful Transform run recalculates the quality scores for the table it wrote to.

Question 13

Are file uploads treated differently from live databases?

Accepted Answer

In two ways. They need no network configuration - no IP to whitelist, nothing to open in a firewall. And they are excluded from the daily background refresh, since a file does not change on its own; their scores update when the file is connected or the integration is edited. Everything else, including quality grading and AI descriptions, works the same way.

Question 14

What happens to dashboards and pipelines if we delete an integration?

Accepted Answer

You are shown what depends on it first. Each integration card carries a row listing how many pipelines, dashboards and chat sessions are built on that source, and deletion shows those counts before it proceeds. Deletion is hard-blocked only while a Transform run is actually in progress. If you revoke someone's shared access and they have built assets on that source, you are warned and must confirm explicitly, because those assets are deleted with the access.

Question 15

Which sources can Edilitics connect to?

Accepted Answer

Relational databases, cloud warehouses, NoSQL stores, flat files, Google Sheets and Redis - 24 live connectors in total, covering the common cases such as PostgreSQL, MySQL, SQL Server, Snowflake, BigQuery, Redshift, Databricks, MongoDB and SAP HANA. Warehouse connections vary slightly in what they need: BigQuery uses a service account key, while Snowflake and Redshift use hostname, database and credentials.

Question 16

Do we need to move our data into Edilitics to use the other modules?

Accepted Answer

No, and there is nothing to migrate at any point. Transform reads from a connected source and writes its result into a database you choose. Visualize and AskEdi query connected sources directly. Because a connection is a pointer rather than a copy, cancelling leaves your data exactly where it already was.

Question 17

Who can see an integration by default?

Accepted Answer

Only you. An integration you create is private until you share it. Workspace admins are the exception - they have full access to every integration in the organisation regardless of who created it, and can use, edit, delete or share any of them without being added explicitly.

Question 18

Will profiling slow down our production database?

Accepted Answer

No. Above 16,600 rows the profiler works from a random sample of that size rather than reading the full table, which is specifically to avoid loading your source. It runs server-side against your database, reads structure and statistics, and moves no data to calculate the scores. There is no agent to install and nothing runs continuously against your source - profiling happens on connection, on a manual refresh at most once a day, and on the daily background job for sources that have gone stale.

Question 19

Does Edilitics write anything to our database?

Accepted Answer

Not from Integrate. Connecting tests that the source is reachable and that the connection is secure; it never runs anything that changes your database. The only module that writes is Transform, and it writes to a destination you explicitly choose - which can be a different database from the one it read from.

Question 20

Can we try this before connecting our own data?

Accepted Answer

Yes. Sample data is connected the moment you sign up, so you can open quality grades, read AI-written column descriptions and see the readiness score without connecting anything of your own. The evaluation runs 14 days and needs no card.

Question 21

What if our column names are meaningless codes?

Accepted Answer

That is the normal case and it is what the descriptions are for. The AI writes a plain-English description for a column from its name, type and statistics, and your industry sector if you allow it - so a column called amt is read as a transaction amount in finance and a dosage amount in healthcare. Where the AI guesses wrong, someone on your team edits the description and marks it validated, and from then on every module reads your wording rather than the cryptic column name.

Question 22

Can we connect a warehouse we already run?

Accepted Answer

Yes, and that is the intended path if you have one. Snowflake, BigQuery, Redshift and Databricks all connect as sources - Edilitics reads from your warehouse rather than replacing it. Nothing is migrated out of it and no second copy is created.

Question 23

Who is allowed to change or delete a connection?

Accepted Answer

The owner and workspace admins. Shared users can build on a connection but cannot edit its credentials, delete it, share it onward, or trigger a refresh. Workspace admins have full access to every integration in the organisation regardless of who created it, so a connection does not become stranded when the person who set it up is away or leaves.

Question 24

What happens to our data if we cancel?

Accepted Answer

Nothing, because it never moved. Edilitics holds the shape of your data and not its contents - column names, types, quality statistics and the descriptions your team approved. Your rows are exactly where they were before you signed up, and there is nothing to migrate back.

Question 25

Is this useful if our data is genuinely in bad shape?

Accepted Answer

That is the case it is built for. Grading tells you specifically which columns are weak and why - blanks, low variety, or values in the wrong format - rather than leaving you to guess. From there Transform fixes them by pointing and clicking, and the grade updates after every step so you can see the improvement as you work. Starting from a poor grade is normal; the point is knowing it before you build a report on it.

Question 26

What happens to a spreadsheet when we upload it?

Accepted Answer

It is virus-scanned, encrypted, and stored in isolation before anything reads it, then put through a pass of basic tidying so it can be profiled reliably. Column headers are trimmed and normalised, invisible characters and Excel error tokens such as #N/A and #REF! are cleaned up, empty strings become proper nulls, entirely blank rows and columns are dropped, unnamed columns get real names, duplicate header names are made unique, and nested structures are flattened. Date and number columns are then detected and typed. CSV, Excel, JSON, Parquet, Avro, Feather, PDF and Pickle all go through the same path.

Question 27

Does the cleanup change our numbers or guess anything?

Accepted Answer

No, and that boundary is deliberate. The upload pass is hygiene only - it never fills in missing values, never removes duplicate rows, never forces a column into a type it does not fit, and never changes the number of rows in a way that assumes what your data means. Anything that needs a judgment call about your business belongs in Transform, where you make it yourself and can see exactly what it did. So an upload leaves you with the same data, only readable.

Question 28

Why do our spreadsheet columns come through with different names?

Accepted Answer

Because raw spreadsheet headers frequently break downstream queries. A header exported from Excel often carries a hidden byte-order mark, trailing spaces, or characters that are not valid in a column name, and blank headers arrive as Unnamed: 0. Those are normalised into clean, consistent names on upload, with collisions made unique. Nothing about the values changes, and if a name reads oddly you can describe the column in your own words in the Metadata Viewer - every module then uses your description rather than the header.

Connect your data sources.Ask them anything.

Your First Connection

Connect a source. Everything else happens on its own.

Open the firewall

Connect the source

Profiling runs on its own

Descriptions get drafted

Your team locks them

Before It Reaches A Deck

Find the broken column before it reaches a board deck.

DQ Scoring Formula

A–F Grade Scale

What Gets Profiled Per Column

Whether Your Data Is Ready

Your AI stops guessing what your columns mean.

Undocumented

AI-Generated

Human-Validated

It Knows Which Industry You Are In

Your Definitions, Not Its Guesses

Your team's definitions, locked against the AI.

Where Your Data Stays

Your records never move. Your credentials never leak.

Your Records Never Move

Credentials Encrypted Per Workspace

We Test, We Never Touch

Uploads Scanned Before Anything Reads Them

Three Privacy Modes

Decide what the AI sees. Once.

Private Mode

Balanced Mode

Full Context Mode

Sharing A Connection

Give your analyst access without giving them the password.

Domain Guard

Credentials Stay Hidden

Someone Can Always Take Over

When A Source Changes

Know what breaks before you change anything.

Columns Change Without Telling You

You Can See What Is Running

See What Depends On It First

Everything you need to know before you decide.

Know what your data is worth before you trust an answer.

Connect your data sources.
Ask them anything.

Connect a source. Everything else
happens on its own.

Find the broken column
before it reaches a board deck.

Your AI stops guessing
what your columns mean.

Your team's definitions,
locked against the AI.

Your records never move.
Your credentials never leak.

Give your analyst access
without giving them the password.

Know what breaks
before you change anything.