Questions and answers
The full list, grouped. The short version: it is free, it needs no account, it runs on your device whenever your browser can, and nothing is stored. If something is not working, help is the page for that.
Jump to: The basics · Privacy · Browsers and devices · Files and limits · Results
The basics
What is Zengist?
Zengist is a free web app that turns documents, pictures, links and text into one brief. It runs on your own device whenever your browser can, and every result says where it ran.
Is Zengist free?
Yes. There is no account, no trial and no paid tier. Most runs happen on your own device, which costs nothing to serve; the cloud fallback is capped per day so it stays free too.
Do I need an account?
No. Open the page and drop a file. There is nothing to sign up for and nothing to install.
Does it work on a phone?
Yes. Safari 26 on iPhone and iPad can download a small private model and run on-device. Most Android phones use the cloud fallback. Either way the page works the same; the label under the brief says which happened.
Can I install it as an app?
Yes. It is a progressive web app: use your browser's Install or Add to Home Screen option and it opens in its own window, with the same on-device engines.
Privacy
Do my files get uploaded?
Not when the run happens on your device, which is the default whenever your browser can do it. If a run uses the cloud fallback, the extracted text (never the file) is relayed to the model and dropped. Nothing is stored on our side.
How do I know where a run happened?
Every result carries a label: on-device with a green mark, or cloud with a blue one. You can also pin an engine in settings, including a setting that never uses the cloud.
Is anything saved after I close the tab?
No. There is no history and no recent list. A brief lives in the tab's memory and is written nowhere, so a shared computer keeps no trace of what was summarized. Your settings are kept in the browser only if you allow functional storage in the cookie notice.
What does the site store about me?
One first-party cookie that records your choice in the cookie notice, your settings in the browser if you allow them, and, while analytics is on, two analytics cookies that count visits. Turn analytics off in the cookie notice and nothing is counted. The full policy is on the privacy page.
Can I keep everything off the cloud?
Yes. In settings, choose Chrome built-in AI or the private model instead of Automatic. If the chosen engine is not available, the run stops and says so rather than falling back.
Browsers and devices
Which browsers run it on-device?
Desktop Chrome and Edge 138 or newer use the browser's built-in model after a one-time download by the browser. Any browser with WebGPU, including Safari 26 on Mac, iPhone and iPad, recent Firefox, and Chrome on most hardware, can download a private model instead. Everything else uses the cloud fallback.
What does Chrome's built-in AI need?
A desktop operating system, about 22 GB of free disk space for the browser's model download, and either a GPU with more than 4 GB of memory or 16 GB of RAM. The browser downloads the model once and shares it across every site that uses it.
How big is the private model download?
About 1 GB on a computer and about 350 MB on a phone. It is always opt-in: the size is shown and nothing downloads until you agree. The download is cached, so the second run starts immediately.
Does it work offline?
Chrome's built-in model works without a connection once the browser has it, and the page itself is cached. The private model path may need a connection for its first run on a device. Links and the cloud fallback always need one.
Files and limits
What can I drop in?
PDF, Word (DOCX), plain text, Markdown, CSV, HTML, JSON, logs, subtitles, images and screenshots, pasted text, and links to public web pages. Up to 20 sources and 10 links per run, 50 MB per file.
Does it read scanned PDFs and photos?
Yes. Text recognition runs in the browser, on up to 30 pages per scanned PDF and on any image. Clean, straight scans and screenshots read well; blurry or tilted photos and handwriting read poorly.
How long a document can it handle?
On your device there is no length cap: long material is split on paragraph boundaries, reduced to notes, and combined. The cloud fallback accepts about 240,000 characters per run, roughly 40,000 words.
What is not supported?
Password-protected PDFs, .doc files, encrypted archives, and web pages that need a login or a paywall. Save old formats as .docx, remove PDF passwords, and paste text from pages you are logged into.
Results
How good is the on-device summary?
Chrome's built-in model is tuned for summaries and does well on everyday documents. The small private model is fine for short material and weaker on long, dense text. For a big pile, the cloud model is the strongest option, and you can pick it in settings.
Can I ask for something other than a summary?
Yes. Type the task: rewrite this plainly, list every deadline, turn these notes into an email, compare the two contracts, what do I owe and by when. The output keeps the same calm structure.
What do Short, Medium and Long change?
How much of the middle you get. Short is a paragraph you can send on; Medium suits most documents; Long is for contracts and reports where every clause matters.
Can I trust the brief?
Treat it as a fast first read. It can miss things and it can misread a scan. Check anything that matters, especially amounts and dates, against the original before you act on it.
Why did a run take longer than usual?
First runs on a device include a model download or the browser's own setup. Scanned pages and images add a few seconds each for text recognition. Very long material is read in parts and then combined, which takes proportionally longer.
Still have a question?
Read how it works for the engines in depth, features for the full list with limits, or the comparison if you are choosing between tools. Questions, bugs and ideas go to the contact address on the privacy page.