How Zallixormind works

Last updated 16 September 2026

Not a legal document. This is a plain explanation of what happens when you send a message, what the labels mean, and where the limits are.

1. Where your question goes

Every message is checked in your browser first. If it is something that can be worked out rather than guessed — a sum, a word count, a unit conversion, a question about Zallixormind itself — you get the answer immediately and nothing is sent anywhere.

Everything else goes to our server, which passes it to an AI model and streams the reply back.

2. The label above each answer

Every reply is tagged with whichever engine produced it, and the coloured bar to its left matches:

This is deliberate. You should always be able to tell whether an answer was calculated on your device or generated by an online model. The first four labels above never send your message anywhere.

3. What the local engine handles

These are computed, not predicted, which means they are correct rather than probably correct:

• Arithmetic — including brackets, powers, roots, logs and trigonometry
• Equations — linear and quadratic, with the working shown
• Percentages — discounts, tips, increases, percentage change
• Statistics — mean, median, mode, range, variance, standard deviation
• Exact large numbers — 2^200 and 100! are computed precisely, not rounded
• Fractions — 1/3 + 1/6 returns 1/2, not a rounding error
• Primes and factors, combinations and permutations
• Unit conversion — length, mass, volume, time, data, temperature
• Text utilities — word and character counts, case changes
• Single-word synonyms from a built-in list
• Questions about Zallixormind and Zallixor Team
• Dates and times — today's date, days until or between two dates, the weekday for a date, age from a birthdate
• Opposites and definitions — antonyms and single-word definitions from a built-in list
• Number bases — decimal, binary, hex, octal and Roman numerals, in both directions
• Encodings — base64, ROT13, Caesar shifts and URL encoding, all reversible
• Word puzzles — palindromes, anagrams, reversal, letter counts, reading time
• Physics and chemistry — named constants and standard formulas, including Newton's second law, kinetic and potential energy, the ideal gas law, Ohm's law and Coulomb's law
• Numerical methods — root-finding, numeric integration and ODEs, labelled as estimates rather than exact answers
• Matrices — eigenvalues, eigenvectors and rank for 2×2 and 3×3
• Pasted tables — per-column statistics, correlation, sorting and filtering
• Attached images — dimensions, file size, format, dominant colours, transparency, camera and lens, when it was taken, exposure settings, GPS location, and QR or barcodes

Zall Vision measures images; it does not look at them. Telling you what is in a picture needs a vision model, and the smallest useful ones run to hundreds of megabytes over your data connection. Zallixormind does not download that, so it says what it can measure and is honest about the rest.

Images stay on the device you sent them from. They are held in this browser's storage, not in your account, so unlike your chats they do not appear when you sign in elsewhere. Clearing your browser data removes them for good.

The local engine is deliberately cautious. If a question only looks like maths — "half of my team quit", "what is the average salary in Nigeria" — it passes it to a model instead, because a bare number would be the wrong answer.

4. Thinking mode

Switching on Thinking routes your question to a reasoning model that works through the problem in steps before answering. You can expand Thought process to read that reasoning.

It is slower and better suited to problems with several stages. For ordinary questions the default is quicker and just as good.

5. Choosing a model, and limits

You pick which online model answers, from the name at the top of the screen:

• Fast — quick answers for everyday questions.
• Thinking — reasons through harder problems in steps, and shows that reasoning.
• Open — open-weight models, slower but independent.

Our server decides what actually runs, and if it has to use something other than what you picked, the label above the reply says so rather than the request you made.

There is no per-day message limit at the moment. If that changes, the top bar and the sign-in page will both say what the limit is before you hit it. Anything the on-device engines answer never counts towards any limit, however many times you use it.

Local mode, in Settings, stops the app calling any model at all. Anything the on-device engines cannot answer is held as Queued and sent the moment you switch Local mode off.

6. When it gets things wrong

Language models produce text that reads correctly. That is not the same as being correct, and the difference is invisible from the outside.

Expect it to occasionally invent a statistic, a citation, a quotation or a link that looks entirely real. Verify anything you intend to act on, and treat medical, legal and financial questions as a starting point for a conversation with a professional rather than an answer.

Answers labelled Zall Engine are the exception. Those are calculated.

7. Where your chats live

In your account, not just on the device you typed them on. Chats are saved to Google Cloud Firestore under your account identifier and synced back whenever you sign in, which is why the same history is there on your phone and your laptop.

Security rules stop any other account from reading them. Because they sit in our Firebase project, though, we are technically able to access them through Google's administration tools — we do not read chats as a matter of course, and would only do so to investigate abuse or a fault, or where the law requires it. If that is not a trade you want for a particular conversation, don't have it here.

Deleting a chat removes it from your account. Deleting your account removes all of them in the same action.

If keeping a conversation matters, copy it out anyway. Every reply has a Copy button, and Settings can export your whole history.

Anything you type that is not handled by the local engine is still sent to a third-party AI provider (Groq, DeepSeek or Hugging Face) to generate a reply. Don't paste passwords, card or account numbers, ID numbers, or anything else you wouldn't want to send over the internet to a company you don't control.

For the formal detail, see the Privacy Policy and Terms of Use.