New: article conversations
At /capture, a reader can send an original article URL and a question, keep a private conversation link, ask follow-up questions and receive replies in the same thread. Existing active shared editions can be returned immediately. New research still needs owner preparation and reply; this workflow does not make anonymous AI calls or log into Camden.
Only short questions/replies and resource references are stored in these conversations. They expire after seven days; expired conversation rows are cleared on the next new request, with fifteen new requests across the demo per UTC day and twelve messages per thread. These request limits are separate from the existing fifteen new sharing links per page per day.
New: capture and discuss
Open the new RET capture studio. Paste a previously captured URL, choose a headline from the dated sample, or bring a readable desktop Chrome page through RET Page Lens. Original source text stays separate from DeepSeek commentary. AI commentary starts collapsed beside each passage: read the article first, then expand individual comments. Saved editions are indexed by their original article URL. Anyone can paste a URL to find an active shared edition, without a new AI call. Private or expired editions remain owner-only; lookup never renews a link. Camden/NewsBank account integration is not connected, and its content is not cached by this index. Sharing links use the existing fifteen-new-links-per-page-per-UTC-day limit and last seven days.
Reader comments are queued for owner approval; recipients need no login. The Chrome extension reads on click and does not take cookies or passwords. Font names and measured sizes are retained, with disclosed fallbacks; proprietary font files are not copied. General capture images/videos remain remote and can fail or change.
The new flow passed 71 automated tests and a production build. File import and the new real DeepSeek path still need a live user test; the 24-article visual benchmark is not complete.
Read EveryThing · 17 October 2026
One link.
Facts, critique and discussion.
A PWA for original educational pages, with DeepSeek behind the scenes and no sign-in for recipients of sharing links.
The Times critical edition
Open the saved Times critical edition as owner, then create a sharing link. This edition retains the user-supplied article text, images, captions, media positions and a newspaper-style layout, with AI commentary beside the relevant material. It is separate from the public-source generator described below.
The exact one-minute Times film played in our desktop Chrome test. The separate PlayCulture tutorial is also included. Video playback requires internet and can depend on provider/device restrictions. The original Times font references are retained, with a bundled Tinos fallback tuned to the source’s fallback metrics; the page identifies which fonts loaded. Live advertising is replaced by labelled slots.
Save HTML exports the article, images, fallback fonts and local comments. Shared links use the same fifteen-new-links-per-UTC-day limit and seven-day expiry. This new edition still needs a physical Android acceptance check.
Public-source generator workflow
- Open the app and choose Create & discuss. Sign in as the owner if prompted.
- Paste the original newspaper URL and choose Create educational page.
- The server researches accessible public sources, checks source relevance and creates a new educational account. It separates sourced facts from AI critique, commentary and discussion questions.
- On the generated page, add your own commentary if you wish. Choose Create sharing link, then copy it to friends or family.
Source text is transient research input. Library credentials, imported reading copies and personal notes are not uploaded by this workflow. It does not log into Camden on readers’ behalf or reconstruct a full paywalled article.
How sharing works
Fifteen new links per page per UTC day. Each link lasts seven days. The owner can revoke all current links for a page. Revoked links still count toward that day’s limit. Repeating a failed request with the same request identifier returns the same link rather than consuming another slot.
This limits new links, not people: an existing link can be forwarded and reopened. Screenshots and saved copies cannot be recalled. Shared pages ask search engines not to index or archive them; that request is not a secrecy guarantee.
Recipients need no ChatGPT, newspaper or library login. Visits do not make AI calls. To bound provider spending, creation allows at most ten research attempts per UTC day, including unsuccessful research attempts. There are no automatic top-ups or credit purchases.
What the AI checks—and what it cannot promise
Source identifiers must exist, and every factual point must have a supporting passage matched in a fetched public page. Supporting passages are discarded rather than republished. The app also checks for long verbatim overlaps and rejects unsupported, overlong or overly close drafts. These automated checks are useful safeguards, not proof that every claim is correct or that a particular use qualifies for a copyright exception.
For video links, YouTube metadata establishes the title and named uploader. A related video remains labelled related; a working metadata endpoint does not prove that it is the newspaper’s exact edit or that playback works in every country. A missing or blocked result stays missing. No footage or link is invented.
AI critique is labelled opinion and should discuss the available evidence, limitations and counterarguments. Your own commentary is labelled separately. The public-source generator uses an independent broadsheet style. The separately identified Times critical edition uses the supplied original source presentation and attribution, with added AI commentary.
Phone and offline behaviour
Keep the existing installed PWA and app address. Android Chrome: refresh the app online; install through the browser menu if needed. iPhone Safari: Share → Add to Home Screen. Creation and current shared links need the network. The existing personal library remains device-local and has its separate offline/backup controls.
Source and hosting
Download the MIT source package. It includes the backend, database migration, reader, tests and the supplied Times critical-edition source. No provider key, library card or PIN is bundled. The application code licence does not relicense the article, masthead, photographs or video; see THIRD_PARTY_NOTICES.md. Tinos carries its included open font licence.
The live build uses the existing Sites Worker and a durable database. The portable research/rendering rules are separated from hosting. An AWS move needs equivalent owner authentication, a database with atomic writes, private secret storage and HTTPS; deploying this version does not migrate it to AWS.
Server settings: DEEPSEEK_API_KEY or DEEPSEEK-API-KEY as a Secret, and READER_OWNER_EMAIL for the creator account. Keep provider secrets out of browser code and source control.
Save or share this guide
This guide works offline; its external links need internet. Save the HTML file, then attach that file to share it. Native file sharing depends on the browser.