Questions

Before you connect anything

The questions worth asking of any tool that wants access to your database, answered specifically rather than reassuringly. If yours is not here, ask us directly.

Do you train an AI model on my data?

No. Nothing you connect, upload, or write is used to train or fine-tune a model. Business context you add to Knowledge Hub is stored in a private index and retrieved at question time, so an answer is only ever grounded in text you can go and read. There is no training step anywhere in the product.

Can it write to, change, or delete anything in my database?

Not on the read path, and not by accident. Every generated query is parsed before it runs and rejected unless it is a single plain SELECT: no INSERT, UPDATE, DELETE, DROP, ALTER or TRUNCATE, and no second statement smuggled in after a semicolon. Writes exist only as a separate, deliberately awkward feature: the connection owner must opt in per connection, only a Data Scientist user can propose one, and it goes through an explicit approval step. Leave it off and the connection is read-only, enforced in code rather than requested in a prompt.

What can I connect?

PostgreSQL, MySQL, Snowflake, and spreadsheets (.csv, .xlsx, .xls). You can also load a sample dataset with one click if you want to see how it works before pointing it at anything of your own.

What do you actually store?

Connection credentials, encrypted with AES-256-GCM (envelope-encrypted through AWS KMS when that is configured) and never logged in plaintext. Your questions, the SQL that was generated, and the results of a conversation, so chat history reloads without re-running queries. An audit record of every question: who asked, what SQL ran, how many rows came back, and any error. Uploaded spreadsheets are copied into our storage because they have to be, which is the one case where your data lives on our side rather than only in your database.

How much of my schema is sent to the AI model?

Only the tables relevant to the question you asked. Schemas are embedded into a vector index and the matching subset is retrieved per question, rather than pasting your entire schema into every prompt. Column values are not sent as part of the schema; data profiling, which samples distinct values to improve accuracy, is opt-in per connection and off by default.

Can it leak sensitive values into an answer?

Results are scanned on the way out and card-number-shaped values are redacted before anything is displayed, with a Luhn check so an ordinary sixteen-digit ID is not mangled. You can also mark specific columns as off-limits on a connection, and a query that references one is rejected before it runs.

Who on my team can see my queries and answers?

Personal connections and conversations are yours alone. Sharing is explicit: a connection can be shared with a team or an account, and only then can other members query through it. Platform admins can see audit records, which is the point of an audit trail.

What happens when the AI does not know the answer?

It says so. When there is not enough context to answer, the instruction it is given is to say that plainly rather than guess, and when a query fails after its retry budget you get the error and the SQL it tried rather than an invented number. Every figure in an answer comes from a query you can read.

Do I need to know SQL?

No. Questions are asked in plain English. The generated SQL is always shown, never required, which matters more than it sounds: it is how someone on your data team can check the answer without re-deriving it.

What does the free plan include?

One connection, fifty questions a month, one dashboard, and the shared platform AI model, so there are no API keys to bring. No credit card is required to start. Sample datasets do not count against your connection limit.

What happens to my data if I cancel?

Your account keeps working to the end of the paid period and then drops to the free plan. Deleting a connection removes its stored credentials, and deleting your account removes your connections, conversations, saved work, and uploads. Your own database is never modified by any of this: we only ever read from it.

Connecting from your own network? See how MyDataTalk reaches your database.