FAQ

The questions people actually ask

Can I get a first look at my data with AI, without uploading it anywhere?

Yes, that is exactly what Totsum Data Explorer is for. The language model and the SQL engine run on your own machine, so your files, questions and results are processed locally. Results are a first look at your data, not an analysis. You shortlist what deserves a closer look, and you hand it on in a follow-up report that is built to be verified. Totsum can work on a computer with no network connection. That includes a machine with no internet access at all, which can stay disconnected.

Can a question use more than one file?

Yes. A question can span several imported files. Totsum looks for the relevant tables and joins them, and there is no separate preparation pipeline to build first. Questions that span files are among the harder ones to get right, so check the preview of the rows behind the answer with particular care.

What does “Ingest” mean?

“Ingest” is the data world’s word for bringing a file into a tool. You drop a file in, and Totsum reads it into tables that you can question. The file itself stays where it is, unchanged.

It also names the first of the app’s three panes, Ingest, Explore and Validate, which form the same loop that the home page draws.

What is a “private beta”, and how do I join?

A beta is the step before an app goes on sale. A small group of early users runs it first, free of charge, and what they run into shapes the finished product. We call it private because you join by applying rather than by downloading. For Totsum, you apply with your email after checking that your machine meets the hardware requirement. We send one confirmation email. Places are limited, so you’ll hear from us only if you’re selected. Selected testers receive an invitation with everything needed to install and start.

Where is Totsum sold?

Totsum is sold in the European Union only. Other regions may follow. If you are elsewhere, applying for the beta with your country tells us where the demand is.

Why not just run a local AI model myself?

Tools that let you run a model on your own machine give you the model, and nothing else. Totsum is a finished product built around one, and the model is the smaller part of it.

What you would miss is everything that surrounds the model: the way your question and your data are presented to it, the checks that run on its work before an answer is shown, and the query flow that lets you read what was actually computed. That is what makes the answers usable rather than merely possible.

Around that sit file import for CSV, multi-sheet Excel, JSON and Parquet, a SQL engine, charts and the follow-up report. With a raw local model, you would assemble and validate all of that yourself.

Which model does Totsum use?

Totsum ships with a local open-weights model, Google DeepMind’s Gemma 4 (12B, under the Apache License 2.0), which weighs about 7 GB. The installer downloads it once, during setup, from our own download server in France, and that is the only place the installer connects to. The model then runs in memory on your own machine, and that local model is what answers your questions.

How can I check that Totsum works with no network connection?

Once the one-time model download is done, switch off Wi-Fi or unplug the network cable, and use the app normally. Importing files, asking questions and building reports all work the same way, because everything runs on your machine.

Can you trust AI-generated SQL?

Not blindly. Like any AI, Totsum will sometimes get a query wrong, so the app gives you the means to check it. Every answer comes with a preview of the rows behind it, and, unless the query is too complex, with a query flow that describes in plain language what was read, filtered, grouped and calculated. The generated SQL itself is one click away.

For the results you keep, the check is done by a person: what you shortlist leaves the app as a follow-up report, and your reviewer records the verdict on each result.

Who is “the reviewer”?

The reviewer is the person you choose to check the numbers before anyone relies on them: a colleague, your accountant, an analyst, or you, if that is your competence. The reviewer is not an AI, and not someone Totsum supplies.

How long does an answer take?

Minutes, not seconds, on most qualifying machines. The model runs on your own computer, and the time depends on your hardware, on your files and on the question. A long wait does not mean that your question was wrong.

You do not have to watch the wait. Ask the question, do something else, and the answer is there when you come back, with its chart and the rows behind it.

If the same machine becomes much slower than it used to be, see the troubleshooting entry below.

Accuracy and trust

How accurate are the previews?

Like any AI, Totsum will sometimes get an answer wrong. Over enough questions that is a certainty, and it happens more often on some kinds of data and questions than on others. That is why the results are called a first look at your data, not an analysis, and the app says so on screen.

The check is part of the product. Every answer comes with a preview of the rows behind it and, unless the query is too complex, with a query flow that you can read; the generated SQL is one click away. The results you keep go into a follow-up report, where your reviewer records Valid, Invalid or Inconclusive against each one.

With use, you build a feel for your own data, and you learn which questions and which corners of your files deserve the closest look. An answer that turns out wrong at review is rarely wasted: understanding why it was wrong usually teaches you something real about the data, such as what a column actually holds, or which assumptions the dataset makes without saying so.

What kinds of questions work best?

Concrete questions that the data can answer work best. Totals, averages, trends, comparisons, rankings and distributions over the columns that your file actually contains are the typical cases. Ambiguous phrasing, such as “the best customers”, works better when you define it, for example as “top 10 customers by 2025 revenue”.

Can I help it understand my data?

Yes. Import a data dictionary, which is a Word, PDF or Markdown document describing your columns and business terms, and Totsum uses it to interpret your schema.

Files and formats

Which file formats are supported?

• CSV (and similar delimited text)
• Excel workbooks, including multiple sheets
• JSON, nested or flat
• Parquet

How big can my files be?

The SQL engine handles files with millions of rows on compatible hardware. The practical ceiling is set by your system memory and free disk space rather than by the GPU. Very large files mostly cost import time and disk space.

Privacy

Where is my data processed?

Totsum processes your datasets, questions, generated queries and results on your machine. The model, the SQL engine and the chart renderer all run locally, and the app has no account, no sync, no telemetry and no update check. It can work on a computer with no network connection. The installer downloads the AI model once during setup. On a machine with no internet access, the model files can be provisioned by hand instead. See the privacy policy for the precise statement of what the software does.

Do I need an account?

No. You need no account for the trial, and none after buying. Licenses are verified locally. The trial does need an email address, because the free trial key is sent to it once you accept the license agreement.

What will you email me?

Your license key, and then two kinds of message. There is no newsletter.

The first kind is a notice about your license: a security problem that has been fixed, the end of your version’s support period, or a change that affects your license.

The second kind is a request for your feedback on how the app serves you. Because the app reports nothing back to us, this is how we learn. It comes at most twice a year, or as often as the programme needs during a beta. Replying is voluntary, and a reply saying “stop” ends the requests.

Trial keys receive the key and the security notices only. The privacy summary says the same in one screen.

Troubleshooting setup

The model download is slow or interrupted

The model is ~7 GB, so on a slow connection the download takes a while. The installer resumes partial downloads, so leave it running or restart it. You can also pre-provision the model offline.

The app says my hardware is insufficient

See the hardware compatibility guide. If your GPU has less than ~12 GB of VRAM, or if your Mac is below the requirement of an M3 Pro chip or better with 18 GB of unified memory, the model cannot run with usable performance.

Answers are much slower than they used to be

A few minutes per answer is normal, as How long does an answer take? explains. If the same machine has become distinctly slower than it was, check that no other GPU-heavy or memory-heavy application is running, and check that Totsum is using the dedicated GPU. On Windows laptops with two GPUs, force the dedicated one in your GPU driver settings.

Something else is wrong

See Report a bug.