Local Summarizer

Right-click selected text to summarize with a small local ML model. Nothing you select ever leaves your browser.

Local Summarizer turns any selected text into a short summary — generated entirely on your own device, with nothing ever sent to a server. HOW IT WORKS Select text on any webpage, right-click, and choose "Summarize." A summary appears instantly in the side panel, generated by a small instruction-tuned language model running fully in your browser (via WebGPU when available, falling back to WASM/CPU otherwise). No text you select — or anything else about your browsing — is ever transmitted anywhere, for any model. There are no developer-run servers behind this extension. Prefer to collect a few passages first? Choose "Add to summarize queue" instead, gather as many selections as you like from different pages, then summarize them all in one go from the Queue tab. KEY FEATURES • Right-click "Summarize" or "Add to summarize queue" on any selected text, on any site. • Fully on-device summarization — your text never leaves your machine, for any model, ever. • Adjustable summary length, with sensible bounds that automatically scale to how much text you selected (so a short selection can't be asked for a 20-sentence "summary," and a long article isn't stuck with a 1-sentence one). • "Resummarize" button to instantly regenerate a summary after changing the length slider or switching models — no need to reselect your text. • Instant zero-download preview while the local model loads or downloads, using a lightweight extractive summarizer, so you're never staring at a blank panel. • Full history of past summaries, with source page title/link, generation time, and which method produced it — remove individual entries or clear everything at once. • Export your entire history to a JSON file whenever you want your own data out. • A model catalog to choose from in Settings — from a fast, lightweight model to a larger one for better-quality output. Each model downloads once from Hugging Face and is cached by your browser for instant offline reuse afterward. • Optional startup behavior settings: load the model only when needed (default), pre-fetch it in the background as soon as the browser starts, or keep it warm in memory the whole time for instant first-use summaries. • Light, dark, or system theme. • Works fully offline once your chosen model has been downloaded once. PRIVACY, IN PLAIN TERMS This extension collects nothing. Your selected text, your summary history, your queue, and your settings are stored only in your own browser's local storage — never transmitted, never sold, never shared. The only network requests this extension ever makes are to Hugging Face's public model hub, to download a model's own weight files the first time you use it — never your text, never any personal data, and never tied to an account, because there is no account. Clear your history any time with one click, or simply uninstall the extension to remove everything it has ever stored. WHO IT'S FOR Anyone who wants a fast, private way to condense long articles, documentation, emails, or any other on-page text — without copy-pasting into a third-party AI tool or chat window. Because everything runs locally, it also works great on sensitive or confidential text you wouldn't want to send to an external API. Note: as with any small (0.5B–1.7B parameter) local language model, output is generally solid but can occasionally include a minor typo or an unusual opening phrase — a deliberate trade-off for running fully on-device instead of relying on a large hosted model.