# What leaves your browser

> Which tools run entirely on your device and send nothing anywhere, which ones call a model, and what is stripped out before they do.

Source: https://www.revquix.com/docs/tools/what-leaves-your-browser
Last verified: 2026-09-04

---

Some tools here never send anything anywhere. Others have to, because they call a model. The split is
not a detail - it decides whether pasting a production JWT or an unreleased salary figure into a box
is safe.

## Tools that run entirely on your device

**Every Money & Tax calculator and every Developer utility runs in your browser.** No request is made,
nothing is stored, and they keep working with the network off.

That covers the whole of both categories - the salary and tax calculators, the invoice generator,
the JSON formatter, the JWT decoder, the hash generator, the base64 encoder, the diff checker and the
rest.

> **Note** - This is why the JWT decoder is safe to paste a real token into
>
> The token is decoded by JavaScript already on the page. It is not sent to us, so there is no server
> log it could appear in and no request for anyone to intercept. The same is true of the hash
> generator and the base64 encoder, including the file modes.

## Tools that call a model

The resume and profile tools are different. They need an AI model, so the text you give them goes to
our servers and then to our AI provider.

**Before it is sent, personal details are replaced with placeholders.** Your name, email address,
phone number, postal address, date of birth and links are removed, and each employer becomes a
neutral label - which is why a report talks about "Company 1" rather than naming your last employer.

That happens on every such request, automatically, with nothing to switch on.

> **Note** - It is also why reports are less specific than they could be
>
> A model that cannot see your employer cannot say "your Infosys line under-sells the migration you
> led". That specificity was given up deliberately, and it is a real cost rather than a free win.

### Two exceptions worth naming

- **Source code is not redacted.** When you ask the assistant about a file in Projects or about a
  practice submission, that code is sent as you wrote it. Redacting code would break it. So do not
  paste credentials into a file and then ask about it.
- **A recipient's name is never sent onward.** In the outreach tool, the person you name is
  substituted out before the request leaves us and put back afterwards, on our own infrastructure -
  see [the outreach tool](https://www.revquix.com/docs/tools/recruiter-outreach-dm).

## Two things people assume are external and are not

- **Code you run** - in the Playground, in Projects and in Practice - executes in a sandbox on our
  own servers, not on a third-party service.
- **Semantic search** runs on a model we host ourselves.

## The authoritative version

This page describes what the tools do. The policy - who the provider is, what the legal basis is,
how long anything is kept - is at [our AI policy](https://www.revquix.com/legal/ai-policy) and the [sub-processor list](https://www.revquix.com/legal/sub-processors). Where the two differ, the legal
pages are correct and this one is out of date.

## Related

- [Exporting your data](https://www.revquix.com/docs/account/exporting-your-data) - Getting a copy of everything.

- [Developer tools](https://www.revquix.com/docs/tools/developer-tools) - The whole browser-only set.
