Distill
Summarize an article, a transcript, or a whole PDF — any length — with a real summarization model that runs entirely on your own machine. Nothing is uploaded, and once it loads it works offline.
🔒 100% private — your text never leaves your devicePaste text below, or drop a file here
.txt, .md, or .pdf — read locally, never uploaded
Summary
Show the key sentences in the source
How it works
Distill runs a real neural summarization model — a quantized build of DistilBART — directly inside your browser tab using WebAssembly. When you hit Summarize, your text is tokenized and fed through the model on your own CPU, and the summary comes back without a single byte being sent to a server. If you drop in a PDF, the text is pulled out locally in your browser too.
How it handles documents of any length
A summarization model can only read about a thousand words at once. Distill gets around that the way a person skimming a long report would: it splits your text into sections, writes a short summary of each one, then stitches those together and summarizes the result. For very long inputs it repeats that reduce step until everything fits, so you can summarize a transcript or a paper that is many times longer than the model's own window.
Load once
The model downloads a single time (~270 MB) and is cached, so later visits start instantly and work offline.
Split & summarize
Long text is chunked into sections and each is summarized on your device, with live progress.
Reduce
The section summaries are combined and summarized again into one clean result at your chosen length.
Trace it back
Distill also highlights the sentences in your original text the summary drew from most.
Why summarize in the browser?
Most summarizers make you paste sensitive text — a private email thread, an internal doc, a medical note — into someone else's server. Distill never does that. The whole thing is one static web page and an open model, so your words stay on your machine, there are no usage limits, and it keeps working on a plane with the wifi off.
Is my text uploaded anywhere?
No. The model runs locally in your browser with WebAssembly, and PDF text is extracted locally too. Nothing you paste or drop in is sent to a server.
How long does it take?
The first run has to download the model (about 270 MB), so give it a minute on a normal connection. After that it is cached. Summarizing itself runs on your CPU, so a short article is quick and a long PDF takes longer while each section is processed.
How accurate is it?
DistilBART is a solid, widely used summarization model, but it is a compact one running quantized for speed, so treat the summary as a fast first pass, not a substitute for reading anything important. It works best on well-structured prose like articles, transcripts, and reports.
What can I summarize?
Paste any text, or drop a .txt, .md, or .pdf file. It is built for articles, blog posts, meeting or video transcripts, release notes, and papers.
Is it really free?
Yes. No sign-up, no account, no limits. It is one of a series of free, privacy-first tools I build in public.
I build tools like this every day.
I'm Harman, a senior full-stack engineer, available for senior or contract work, fully remote. See the rest of the lab or get in touch.