Why "just paste it into ChatGPT" isn't the whole answer
Cloud chatbots are good at summarizing short things you can paste into a chat box. The moment your source stops being a paragraph — becomes a full article, a 40-page PDF, a Word doc, a photo of a page — the workflow starts leaking. You copy-paste and lose formatting; the article is truncated because it was too long; the PDF has to be uploaded, which is fine for a news article and awkward for a contract, a medical letter, or an unpublished manuscript.
None of that means AI summarization is a bad idea. It means the tool should meet the document where it lives — locally, in the format it's in, and preferably without a round trip to a server you don't control. That's now a thing you can do.
What "everything" actually covers
A summarization tool worth installing should handle the whole zoo of formats that reading actually comes in. In practice, that means five buckets:
- Web articles — paste a URL, the tool fetches the page, strips the ads and navigation, and summarizes what's left. Long-reads, essays, blog posts, wiki pages.
- PDFs — reports, whitepapers, research papers, contracts, boarding-pass-length receipts, book-length manuscripts. The text is extracted locally, so nothing uploads.
- Word & text files — DOCX, plain TXT, Markdown, HTML files, RTF. The formats colleagues actually send you.
- Photos of text — a page from a physical book, a whiteboard, a printed handout, a screenshot where the text isn't selectable. On-device OCR reads the words out of the image; then the summary runs on the words. Latin and CJK scripts are both supported, so English, European, Chinese, Japanese and Korean pages all work.
- Videos — and if the "reading" you're avoiding is really a three-hour YouTube podcast, that has its own guide.
Cloud copy-paste vs. dedicated web tool vs. on-device app
Three broad ways to actually do this, honestly compared:
| Approach | Best for | Trade-off |
|---|---|---|
| Paste into ChatGPT/Claude/Gemini | Short web articles you don't mind sharing | Length caps, awful for PDFs, uploads whatever you paste |
| Cloud "summarize" web tool | Quick one-offs on public documents | Uploads the file to a server you don't own; often quality-throttled on the free tier |
| On-device app (Gist&Chat) | Everything, especially private documents | One-time model download; capability ceiling below the very largest cloud models |
The step-by-step: three seconds to a summary
Articles by URL
Copy the article's URL and paste it into the app. The page is fetched, ads and navigation are stripped, and the model summarizes what's left. Works with most modern article sites, including paywalled pages you already have access to in your own browser.
PDFs and document files
Tap the file button, pick the PDF (or DOCX/TXT/MD/HTML/RTF) from your phone's storage or a cloud-drive share sheet. The text is extracted locally before anything else happens. Long PDFs are processed in passes so the whole document gets covered, not just the first few pages.
Photos of text
Take a picture — or pick one from your camera roll — of a page you want summarized. On-device OCR pulls the text out; the summary runs on that text. The physical-book workflow, at last.
Then: ask it questions
Every summary comes with a follow-up chat. "What did the second author disagree with?" "Skip to the section about pricing." "Explain that graph in plain English." The AI still has the whole document loaded, so the answers are grounded in what you gave it.
Why "offline" is a bigger feature for documents than for chat
An offhand chat message and a legal contract have very different privacy weights. When you paste an entire PDF into a cloud tool, you are — no exaggeration — depositing a copy of that document on someone else's infrastructure, subject to their retention policy, their subprocessors, and whichever government asks nicely later. For a news article, fine. For your employment agreement, your child's IEP, an early draft of your novel, your medical results, or a research paper you haven't published yet, obviously not fine.
The on-device version is different by construction — as we get into in what on-device AI actually is. The document is opened, read, and summarized without ever leaving your phone. The airplane-mode test works here just like it does for private chat: turn the network off, run the summary, and observe the document going nowhere.
The honest limits
A guide that doesn't list what a tool can't do is an advert. Some straight talk:
- Very long documents take longer. A 300-page PDF is a lot of text; the model reads it in chunks, and that takes real seconds on a phone. Not minutes — but not the near-instant feel of a two-paragraph web article.
- Scanned PDFs need OCR. If a PDF is a scan of a page (an image inside a PDF wrapper), extract-text won't find any text. Screenshot the page and use the photo path instead, or use a desktop OCR tool first.
- Login-walled articles need you logged in. A URL summary sees what a browser sees at that URL. If the article is behind a paywall or login, the fetched page is the login prompt. Save it as a PDF first ("Print → Save as PDF") and summarize that.
- Compact models < frontier cloud models. An on-device model isn't the largest cloud giant. For summaries, translations and Q&A over documents — the tasks this article is about — the gap is small and the privacy gain is total. For expert reasoning at the very top end, cloud still wins.
Read faster. Upload nothing.
Gist&Chat — the on-device AI that summarizes articles, PDFs, docs and photos of text, then chats with you about them. Free on Android.
- All the formats — URL · PDF · DOCX · TXT · MD · HTML · RTF · Photo (OCR)
- Nothing leaves your phone — verify it with airplane mode
- Ask the summary questions — follow-up chat with the whole document in memory
- 11 languages — interface and summaries
- Free to start — no account, no upload, no ads
Free to download · One-time model download on Wi-Fi · Privacy policy that barely needs to exist
Frequently asked questions
Can I summarize a PDF without uploading it anywhere?
Yes. Gist&Chat opens the PDF on your phone, extracts the text locally, and feeds it to a language model that also runs on your phone. Nothing is uploaded — which is the whole point when the PDF is a contract, a medical record, or a piece of research you're not ready to hand to a cloud service.
What file formats can I summarize?
PDF, Word (DOCX), plain text (TXT), Markdown (MD), HTML files, and RTF — plus any web article by URL, and any photo of text via on-device OCR. Effectively: if it's readable, it can be summarized.
Can I summarize a screenshot or a photo of a printed page?
Yes. On-device OCR reads the text out of the image — Latin and CJK scripts, so English, European, Chinese, Japanese and Korean pages all work — and then the summary runs on that text. Useful for physical books, whiteboards, and screenshots where the text isn't selectable.
Is there a length limit on articles or PDFs?
There is a limit on how much text a language model can hold at once, and long documents are chunked and processed in passes so the whole thing gets covered — not just the first few pages. In practice, book-length PDFs work; you just wait longer for the summary.
Is my document actually private if the AI runs locally?
Yes — structurally, not as a promise. The file never leaves your phone. You can verify it by putting the phone in airplane mode and running a summary: if it still works, the document is provably not going anywhere.
More from gist-ai.net: how to summarize any YouTube video with AI, what on-device AI actually is (and why it works), on-device small language models & why Gemma 4 changed the map, the AI chatbot that works in airplane mode, practice a language with an AI in your pocket, offline ChatGPT alternatives — the real options in 2026, the 8 best podcasts on YouTube, and more free tools — no sign-up, ever.
