How Gauth Works: From Question to Worked Solution

Five things happen between pressing send and reading step one.

Your browser posts the problem to our server. Not to the AI provider. The key that authorises those model calls lives only in the server environment, so it never appears in the page source and cannot be lifted out of your network traffic and spent by someone else.

The server checks what you are entitled to. A first-time visitor gets one full solution. A signed-in free account gets 3 a day. A Plus subscription skips the check.

The server builds the request. Your problem is sent to a large language model together with a fixed set of instructions about how to answer, at a low temperature setting so the same problem tends to produce the same working rather than a different improvisation each time.

The reply streams back. The model produces text token by token, and we forward each chunk to your browser as it arrives. You are reading step one while step four is still being written.

Your browser renders it. The mathematics arrives as LaTeX between dollar signs and is compiled into real notation — fractions stacked, integrals with proper limits, matrices in brackets — and the final answer is split into its own block at the end.

That is the whole pipeline. There is no computer algebra engine underneath and no bank of pre-written answers being matched to your question. The working you read was written for your specific problem, in the moment, by a model that is predicting text. That is what makes it flexible, and it is also exactly why it can be confidently wrong.

Every request carries the same instructions:

Number the steps from one. After each step, say why that move is valid. Put every piece of mathematics in LaTeX rather than flat keyboard characters — typed as x^2/3 a term could be $\frac{x^2}{3}$ or $x^{2/3}$, and rendering it removes the guess. If the problem itself is ambiguous, state the assumption you are working under instead of choosing one silently. End with a clearly marked final answer, and check the result where checking is cheap.

The justification requirement is the one that matters most. A step that states its reason can be audited: “multiply both sides by four to clear the denominator” is a claim you can agree or disagree with. A line of algebra with no reason attached is one you have to take on trust, and trust is the wrong posture to have towards a generated answer.

A photo is read in your browser, converted to an inline image, and sent with the same request as any typed context you add. Images need to be under 4.5 MB, and you can paste a screenshot straight into the input rather than saving it first.

Two practical notes. Typed context alongside a photo works well — “part (b) only, and we have to use integration by parts” will steer the answer. And the model sees only what is inside the frame. If the question refers to a figure, the figure has to be in the shot; when it is missing, the model tends to assume a plausible configuration and keep going rather than tell you it cannot see the diagram.

Once you have an answer you can keep asking about it — why a step is valid, what happens if the sign flips, whether there is a shorter route. The last few turns of the conversation are sent along as context so the model knows what “that step” refers to.

Follow-ups never consume your daily quota, on any plan. Metering the questions that produce understanding would make copying the answer the cheapest thing to do, which is the opposite of what this product is for. They do need an account, because a conversation has to be attached to something.

The first solution is free with no account and no card. After that, a free account gives you 3 solutions a day, unlimited follow-ups, and saved history — including the solution you generated before you signed up, which gets moved across rather than thrown away.

Gauth Plus removes the daily limit. It is $11.99 a month with a 3-day free trial, $31.99 a quarter, or $99.99 a year, all billed through Stripe and cancellable from your account. The pricing page has the full comparison, and the help centre covers changing or cancelling a plan.

It does not check its own arithmetic against an independent engine. It does not know your instructor’s conventions, your syllabus, or which method you are supposed to be practising this week. And it writes with the same confidence whether it is right or wrong, because fluency and correctness are not the same property in a language model.

So the last step of the pipeline is yours: read the accuracy page for the specific situations where solutions degrade and a short routine for checking one before you trust it. If you want the method rather than the answer, the concept guides derive each technique from scratch.