AI data analyst tools: which kind you actually need
Search for "AI data analyst" and you get a list that mixes two completely different products as though they compete. They do not, and picking the wrong kind wastes a fortnight.
Upload tools take a file and answer questions about it. Connected tools query your live database. Everything below is about telling which problem you actually have.
The two kinds, side by side
| Upload a file | Connect to the database | |
|---|---|---|
| You give it | A CSV, Excel file or export | A connection string |
| Freshness | As stale as your export | Live, every time you ask |
| Data size | Whatever fits in a file | Whatever the database holds |
| Repeat questions | Re-export each time | Just ask again |
| What leaves your machine | The file contents | Varies: schema only, or rows too |
| Needs DB access | No | Yes, a connection string |
| Best for | One-off analysis, shared spreadsheets | Recurring questions about live data |
How to tell which you need
Three questions, and they settle it quickly.
Where does the data live right now?If the honest answer is "in a spreadsheet somebody emailed me", use an upload tool. If it is "in our production database", an upload tool means exporting first, and you will get bored of that by the third time.
Does the answer need to be current?A CSV is a photograph. Fine for "how did Q2 go", wrong for "how many signups today".
Will you ask this again? This is the real one. A one-off question is cheap either way. A question you ask weekly is where the export step becomes the whole cost, and where connecting pays for itself immediately.
What connected tools are genuinely bad at
Being honest about this, because the category oversells itself constantly.
They do not know your business. The model can see a column called statusand cannot know that value 3 has meant "churned" since a migration two years ago. It will pick something reasonable and tell you nothing about the guess.
They fail quietly. The dangerous output is not an error, it is a query that runs and returns a confidently wrong number. That is worth understanding properly before you rely on one, which is why we wrote a separate piece on how to catch it.
They are single-player. A desktop client answers your question on your machine. If the same number is needed by ten people every Monday, you want a dashboard, and something like Metabase is a better fit than any client, ours included.
Where FluentDB sits, and where it does not
To be upfront: FluentDB is one of the connected tools, and this is our site.
It is a native Mac app that connects to PostgreSQL, MySQL, SQL Server and SQLite. You ask in plain English, it reads your schema and writes the SQL, and it shows you that SQL before running it. The assistant sees your schema rather than your rows by default, and you bring your own Anthropic or OpenAI key or run a model locally with Ollama. It is a one-time purchase from $54.
Where it is the wrong choice: if your data is in spreadsheets, an upload tool will serve you better. If you need shared dashboards or scheduled reports, you need a BI tool. If you are not on a Mac, it is not an option at all. And if your database is something other than those four engines, Chat2DB covers far more engines and is open source.