BriefFeed Privacy Policy

Website visit and download measurement

The BriefFeed website measures page views and clicks to its five app and extension store listings using Cloudflare Analytics Engine. Events contain only a fixed website route, store platform, button location, registered campaign or broad referral category, device category and country code. They contain no user identifier, IP address, full referring URL, arbitrary query values or account data. Campaign labels stay in this tab’s session storage with a 30-minute expiry refreshed on each website page load so attribution survives navigation. Known automated browsers and visits with bf_analytics=off are excluded from this measurement. Events are retained for three months. A store click indicates download intent, not an installation or successful use. Separate Cloudflare Web Analytics and infrastructure request logs may also measure website traffic.

Optional local Agent access

In builds that include Agent access, you can authorize a local CLI or MCP connection to your saved items, retained history and current browser Insights, including available source excerpts with an 800-character preview and up to 25,000 characters of retained source context per source tree. New authorization also allows saving links, saving retained history to your library, and deleting specific saved items or history. Existing read-only grants remain unchanged until explicitly renewed. Library changes may propagate through your existing device sync; file exports remain read-only. The browser helper reads through the running extension without retaining a separate data file. The Mac helper reads the local Apple library, including data already received through private iCloud sync. These helpers do not upload your library to BriefFeed servers. Your chosen Agent may send the data to its own model or retain results under its own policies. Revocation prevents future access but cannot recall copies already received or undo completed library changes. Exported files are separate copies and must be removed separately. History keeps its original expiry; cloud delivery may be delayed.

What BriefFeed processes

When BriefFeed summarizes a supported source or creates an AI Insights brief, it processes content already loaded in the browser page, including supported X, Bluesky, Jike, Weibo, Reddit, and V2EX post, reply, and comment surfaces in foreground or background tabs before they enter the viewport; the visible title and caption of a Xiaohongshu note detail the user opens, or the matching title and caption delivered in the public page response for a note the user explicitly sends to the Apple client, with the app-provided share text as fallback; current-video YouTube transcript/caption or timeline text and Bilibili subtitle text and chapter markers when available; visible WeChat, DEV Community, explicitly enabled Zhihu article and individual-answer text, or NetEase News article text; visible arXiv paper titles, authors, and abstracts, plus a bounded section-stratified excerpt of main text already rendered on opened arXiv HTML paper pages; visible titles, authors, TL;DRs, abstracts, and keywords on opened public OpenReview paper pages; visible Apple Podcasts or Xiaoyuzhou episode Show Notes; visible Substack, Medium, or WordPress article text; visible Product Hunt product/launch text and comments that entered the viewport; and visible GitHub repository or public profile README text. When browser-history sync is enabled, Insights can also use bounded excerpts from same-day history received from your other devices. BriefFeed does not auto-scroll or paginate to load additional content. The request can include the source id or stable URL, author display label, timestamp, and, for supported long articles, the observed title, public author profile locator and source-account identity evidence, loaded-page or current-video transcript/subtitle/timeline text, current-video chapter markers when present, platform, detected language, summary language, discovery scope, bounded bold-emphasis metadata matching the supplied text, optional thread-layout offsets describing primary bodies and embedded quotes within that same text, and optional loaded-page link metadata or quoted-post context. For Xiaohongshu, its signed access URL and visible image URLs are retained only after an explicit bookmark for reopening and optional external sync; the signed URL is not sent to BriefFeed's hosted summarization service.

What BriefFeed does not collect

• BriefFeed does not ask for, collect, or transmit your social-media passwords.
• BriefFeed does not collect, store, or send cookies, session tokens, login credentials, or account secrets to BriefFeed servers.
• BriefFeed does not upload raw DOM snapshots, complete page snapshots, browser history, full follower/following lists, follower graphs, payment data, or precise location.
• BriefFeed does not auto-scroll or paginate social feeds, open additional social pages, or call platform feed APIs to obtain more posts for AI Insights.
• BriefFeed does not fetch Bluesky app state, DIDs, network responses, or hidden API payloads.
• BriefFeed does not fetch Xiaohongshu feeds or private platform APIs, image text, video, audio, comments, unrelated recommendations, cookies, or account/session data; it reads only an opened note detail's visible title and caption or the matching note record delivered in the public page response for a note explicitly shared through iOS, with the supplied share text as fallback, and visible image URLs only when the user explicitly saves it.
• BriefFeed does not fetch WeChat APIs, account/session data, comments, or media-only payloads. During the summarized-content trial, WeChat entries use the submitted article metadata and generated summary without a separate source-verification fetch.
• BriefFeed does not fetch Substack APIs, subscriber data, email payloads, comments, or paywalled article text that is not loaded in the current page. The public summarized-articles trial may verify the same submitted article through an anonymous public HTML request, as described below.
• BriefFeed does not fetch Medium APIs, member data, recommendation feeds, responses, or paywalled article text that is not loaded in the current page. The public summarized-articles trial may verify the same submitted article through an anonymous public HTML request, as described below.
• BriefFeed does not fetch WordPress APIs, admin/account data, comments, related-post feeds, or article text that is not loaded in the current page. The public summarized-articles trial may verify the same submitted article through an anonymous public HTML request, as described below.
• BriefFeed does not fetch DEV Community APIs, reactions, comments, or account data. The public summarized-articles trial may verify the same submitted article through an anonymous public HTML request, as described below.
• BriefFeed does not fetch Zhihu APIs, initial-state payloads, comments, recommendations, account/session data, or text hidden behind expand, login, or payment controls; it reads only an opened article or individual answer’s rendered title, byline, and visible body. The question provides context; other answers are not combined. The public summarized-articles trial may verify the same submitted article through an anonymous public HTML request, as described below.
• NetEase News summaries use only the opened article’s loaded visible text, excluding comments, recommendations, hidden payloads, cookies and account data.
• BriefFeed does not fetch arXiv APIs, PDFs, citation graphs, or related-paper feeds. On abstract pages it uses the visible title, authors, and abstract. On HTML paper pages it sends at most 24,000 characters sampled across the rendered title, abstract, and main sections, excluding references and appendices.
• BriefFeed does not fetch OpenReview APIs, PDFs, reviews, rebuttals, decisions, ratings, comments, hidden fields, or private forum content. It summarizes only an opened public paper page's visible title, authors, TL;DR, abstract, and keywords.
• BriefFeed does not fetch Product Hunt APIs, hidden app payloads, rankings, votes, reviews, or comments that are not rendered in the page.
• BriefFeed does not fetch YouTube audio, video, comments, recommendations, account data, or Google/Gemini session data. Video text auto-loading is limited to the current video's available YouTube transcript/captions, timeline text, and chapter markers; the browser may use your current YouTube session for that YouTube request, but BriefFeed does not read, store, log, or send those cookies/session tokens to BriefFeed servers. It does not force an English caption track before summarization.
• BriefFeed does not fetch Bilibili audio, video, comments, recommendations, hidden platform payloads, or account data. Video text auto-loading is limited to the current video's available Bilibili subtitle JSON and chapter markers; the browser may use your current Bilibili session for that Bilibili request, but BriefFeed does not read, store, log, or send those cookies/session tokens to BriefFeed servers.
• BriefFeed does not fetch podcast audio, transcripts, RSS feeds, or platform APIs for Apple Podcasts or Xiaoyuzhou; podcast summaries are based on visible Show Notes only.
• BriefFeed does not fetch GitHub APIs, private repositories, account data, hidden repository metadata, or profile text that is not rendered in the page.
• BriefFeed does not create a separate BriefFeed user account.

Optional site access

By default, BriefFeed asks for only the exact HTTPS origin before enabling Xiaohongshu, Zhihu, DEV Community, arXiv, OpenReview, or a detected custom-domain Substack, Medium, or WordPress article site. Chrome users can explicitly turn on automatic site access once, allowing BriefFeed to recognize packaged support on HTTPS pages they open without another per-site permission prompt or notice. BriefFeed does not visit unopened sites and uses either grant only for recognized supported sites—not for general browsing, analytics, cookies, browser history, hidden content, or background crawling. Automatic site access can be turned off at any time.

Optional Apple Mobile Inbox sync

If BriefFeed for Mac is running, the Chrome extension may exchange Mobile Inbox links with it on the same Mac. Mac App Store builds use an HTTP bridge bound only to 127.0.0.1 and restricted to the published BriefFeed extension origin; the extension requests an optional permission limited to http://127.0.0.1/* only when you connect the Mac app, and direct-download builds may use Chrome Native Messaging instead. A random pairing code is displayed for manual entry, never copied to the general pasteboard, expires after two minutes, and provisions a separate connection code stored only in the Mac Keychain and this browser. A one-use signed challenge authenticates the Mac app before the extension sends bookmark data, and the app authenticates each request and response. When the side panel opens or refreshes, and after you save or hydrate supported content, the extension imports pending iPhone/Mac-captured links and syncs resolved Chrome bookmark metadata back to the user's local Apple library. Payloads are limited to link IDs/URLs, optional title, author, bounded capture client/method provenance, note, summary, up to 3 tags, platform/content type, timestamps, status, and UI locale. For a Xiaohongshu share sent explicitly to the iOS app, BriefFeed removes the short link and app boilerplate from the user-visible share text, follows that public short link without cookies to try to read only its matching note title/caption, falls back to the supplied share text when parsing fails, and sends only a URL without signed capability parameters when requesting a summary. For a saved non-article social post, the note can include the complete visible source text up to 25,000 characters, matching X's longer-post limit, so the post remains readable on iPhone and Mac. Other supported forum and community sources can retain up to 4,000 characters of visible source text after an explicit save, including when no summary is generated. BriefFeed does not use either local transport to read arbitrary local files, scan the device, access browser history, cookies, passwords, account credentials, DOM snapshots, hidden post content, full article bodies, or control other apps. If BriefFeed for Mac is not running or no local bridge is available, this sync is inactive.

Apple app link handling and iCloud sync

The iPhone, share extension, and Mac apps can save URLs you enter, paste, or share. Saved items sync through your Apple iCloud account; BriefFeed does not create a BriefFeed account for this Apple sync. When you save a supported URL and the app tries to read metadata or create a summary, it may fetch the public page and send a minimized request to BriefFeed's hosted API. That request can include the saved URL, page title, author label, visible article or post text, limited HTML-derived text for supported articles, platform/content type, summary language, and up to three returned tags. These Apple app requests are used for app functionality, are not used for tracking or advertising, and are not intentionally linked to a BriefFeed user identity.

Hosted summarization service

By default, BriefFeed sends supported summary requests to BriefFeed's hosted backend. The backend validates the minimized request, calls the configured AI provider when needed, and returns the generated summary and tags to the extension or Apple app.

• Request bodies are intentionally narrow and are limited to the fields needed to produce the requested summary.
• BriefFeed does not use post content for advertising or to determine creditworthiness, lending eligibility, or similar sensitive decisions.

Hosted summary usage measurement

When clients send optional technical metadata, BriefFeed measures repeat hosted summary use with a pseudonym derived from a random server-issued subject. Events contain the pseudonym, internal/external classification, client platform/version, content platform, provider and returned summary count. They do not contain raw credentials, IP addresses, source URLs, author names, source text or generated summaries. Events are retained for three months. Known internal traffic is excluded from external retention reports. Clients omitting optional technical metadata, including Firefox without the optional technical-data permission, are not included. Credential renewal or clearing local data can reset this measurement identity. These statistics describe credential instances and server-confirmed outputs, not unique people or proof of reading.

Anonymous hosted access credential

BriefFeed does not require a BriefFeed account. BriefFeed clients request a random server-issued bearer credential for the hosted API. The Chrome and Safari browser extensions store it in extension-local storage, while the native iOS and macOS clients store it in the Apple Keychain. The public-only Agent adapter keeps its separate credential in process memory, without reading private library credentials. The server record for this credential contains only its SHA-256 hash plus a random subject identifier, status, tier, client platform, issued time, and expiry time. BriefFeed uses the credential for API admission, quota enforcement and abuse prevention. When optional technical metadata is enabled, the hosted service also derives a separate pseudonym from its random subject to measure repeat successful summary use. This is not an account and is not used for advertising or cross-site tracking. For hourly and daily abuse-prevention limits, the hosted service also converts the request IP address into a salted, truncated one-way hash. BriefFeed's application rate-limit records do not store the raw IP address; expired hash counters are ignored and removed during cleanup. The raw bearer is not written to backend logs. Any legacy client or install identifier still present in a request body is operational metadata and is not accepted as authentication.

Public author profiles

For identified authors of saved items, supported clients can retain a public display name, profile link, portrait URL, and biography from supported page-visible author surfaces. These fields enrich your author library and can sync through the existing browser-to-Mac connection and your private iCloud library. Biographies are kept separately from saved source text and are not sent to the hosted summarization service. Missing fields do not erase previously observed information. BriefFeed does not crawl author feeds or fetch profiles in the background to fill gaps.

Public summarized content trial

For supported long-form summary requests, BriefFeed can retain the observed article title, validated source-account key and profile link, genuine publication time when available, canonical source link, language, and a reference to the generated summary. This index is separate from your private bookmarks and loaded-content history. The author page can show titles and generated summaries from ordinary reading or after a save; inclusion does not indicate that anyone bookmarked an article. The trial covers eligible WeChat, DEV Community, Substack, WordPress, Medium, X long articles, Weibo long articles and Zhihu articles, including supported custom domains. During the trial, WeChat entries use the submitted article metadata and generated summary, with consistent source-account identity, without independently checking anonymous access or fetching the original again. This does not assert that the original is freely accessible. Other supported sources still require verification: before a summary is listed, the server makes a bounded, credential-free request to that same public article, without cookies or redirect following, and checks the submitted text and author association against the anonymous public response. It does not crawl author feeds or use this fetch to expand the summarization source. For those other sources, protected, paywalled or unverified entries are excluded. All sources exclude context-dependent, expired, blocked or incompatible entries. Verification text and HTML are transient and are not stored. Clients and Agents use the same listing boundary. Catalog rows do not contain source bodies, visitor or collector identities, private notes, save counts, or user-to-article relationships. Author queries send one source-account key, language and pagination cursor, never your bookmark list. Catalog metadata follows hosted-summary retention, while listing eligibility expires after 7 days unless renewed by another eligible summary request. Removing a private bookmark does not remove this independent public catalog entry. For source-level removal, contact hello@brieffeed.app; removal also blocks later catalog readmission. Existing historical cache rows are not bulk-published.

Retention

Visible source text that you explicitly save as a bookmark can remain in local Chrome bookmark storage and, within the applicable text limits for non-article social posts and supported community sources, in your Apple library synced through your iCloud account. Outside saved bookmarks, raw source text is used for the current summary and the short-lived loaded-content retention period, locally and, only with opt-in, in your private iCloud. BriefFeed's hosted cache does not retain raw source bodies. It does retain generated summaries and tags, source URL or ID, author label, source timestamp, language, content and context hashes, optional long-article title and source-account metadata, and provider and model identifiers. These hosted cache records are currently retained indefinitely for cross-user deduplication. To request deletion of a hosted cache record associated with a source URL or ID, contact hello@brieffeed.app. Backend logs must avoid raw source bodies, provider secrets, auth tokens, cookies, and extension storage dumps.

Local browser storage

BriefFeed stores saved bookmark records in the extension's local IndexedDB, including user-saved visible source text, generated summaries and tags, source metadata, sync state, and deletion markers. After an upgrade, existing bookmark records are imported from Chrome extension local storage into IndexedDB and the legacy bookmark envelope is removed. Browser extension storage remains in use for settings such as whether summaries are enabled, preferred UI and summary languages, disabled hosts, user-enabled custom Substack, Medium, or WordPress hosts, Mobile Inbox link metadata, lightweight bookmark revision notifications, the optional Mac connection credential, optional Notion sync configuration, optional NotebookLM target notebook configuration, and developer-tuning values in development builds. The extension retains supported loaded content in a separate local IndexedDB database before AI Insights is enabled, without a daily item-count cap, subject to available browser storage. Loaded content is retained for the last 7 local calendar days, including today, across browser and Apple clients. AI Insights still uses only content loaded today. Expired content is cleared when the extension next runs cleanup, reads, or writes. Bookmarks are retained separately. AI Insights uses today's content, including history received from other devices when browser-history sync is enabled. Synced content qualifies by its original observation time in the receiving browser’s local calendar day and remains subject to expiry and deletion. Insights still requires separate consent. Optional loaded-content sync is off by default. If enabled, bounded source excerpts, URLs, author labels, observation dates, and expiry dates travel through BriefFeed Reader on Mac to your private iCloud for Apple-device search. The same retention period applies. Turning it off stops new uploads and removes this browser’s synced copies after the Mac reconnects; iCloud propagates removals when devices connect. Bookmarks are separate and are not removed. It may also cache recent summary results so repeated viewing of the same loaded post does not require another request. If you enable Notion sync, the Notion token is stored only in this browser and is not sent to BriefFeed servers. NotebookLM sync uses the current browser's Google session and does not store Google cookies or NotebookLM frontend tokens.

Summary display checks

For supported browser, iOS and macOS requests, BriefFeed may omit a short same-language summary that barely reduces the original reading length. The server compares the primary reading length with the actual summary, including when reading a hosted cached summary. Separately supplied quotes, visible discussion context and link-preview prose remain context for evaluation but are excluded from primary reading length; all assembled same-author thread post bodies remain primary content. Optional validated layout offsets let the server exclude assembly labels and embedded quotes from this display-only length calculation without removing any source text from generation or review. These offsets contain no additional source content and are not sent to generation providers. For Xiaohongshu notes, an optional title field repeats only a bounded title already present at the start of the visible source, so the server can exclude it from the body-length comparison while retaining the full source for generation and review. This display check is independent of any earlier pre-generation summary-value decision. For pairs sent to actual-output review, actual translation, script conversion, useful interpretation of visible context and uncertain cases are preserved. An exact repeat of the primary text can be omitted without another model call. Within the short same-language plain-summary scope, standalone outputs and outputs containing quoted posts saving less than 30% of primary reading length are omitted directly, even if rephrasing could offer some benefit. Other sources containing visible context, previews, grounded Xiaohongshu titles or validated thread layouts retain a 10% hard floor and optional actual-output review below 30% savings. Explicit translation modes and shrinking structured outputs retain their existing handling. When actual-output review is enabled, TypeSafe AI's Jev model may compare the bounded visible source and its context with the generated summary to check whether that specific output adds reading value, with OpenAI Decisions API as an optional fallback for unavailable or invalid decisions. The comparison may include the source author's already-visible display label to distinguish attribution from added information. This sends the bounded source, context, author display label when present, and generated summary to the selected decision provider. Low-confidence, useful and uncertain valid decisions retain the summary without asking a fallback provider for a different judgment. This display decision does not delete the stored summary or saved source and does not request a replacement summary. Browser display-cache policy updates require a fresh service check. Current Apple clients can recheck old short plain social summaries during display or enrichment. A local receipt binds the current policy to the retained source, target language and result so an older unreviewed summary cannot overwrite a reviewed skip through sync. Older native app versions and URL-only cache lookups are excluded. Native clients remember skips for the matching source and language; policy changes do not delete saved source content. Existing unreviewed summaries may be hidden after a completed matching review.

Optional summary-value evaluation

When enabled by BriefFeed, an experimental check may send bounded visible social and community source text to TypeSafe AI's Jev model, with OpenAI Decisions API as an optional fallback, to judge whether a same-language summary would add reading value. The evaluated text can include posts, comments, short answers or discussion excerpts, together with their supplied visible question, quoted context and link-preview prose. It can also include the visible profile biography on X/Twitter, Bluesky or GitHub, without collecting the rest of the profile or its posting history. Short X Article previews already loaded in the browser are also included; opened article bodies remain excluded. Long-form articles, structured long answers or notes, podcasts and videos are excluded. This server-side check applies to supported summary requests from browser extensions and native Apple clients, including adding or sharing a link. Observation mode still generates the usual summary; filtering mode may omit a redundant summary while keeping the saved source and link. This check does not add source-URL or author metadata, private account or session data, or cookies. BriefFeed records numeric evaluation and length measurements, not source text, for this experiment. Confirmed skip decisions and reusable language-verification results are stored indefinitely using a content-and-policy hash and numeric scores, without source text. Only TypeSafe results enter those numeric decision caches; OpenAI fallback scores are not cached there. For consistent display across clients, BriefFeed separately retains the first completed show/skip outcome, including outcomes informed by either provider or server length checks, as a source-and-policy hash, disposition and timestamps. This display record contains no source or summary text, author label or URL. They are reused only while the source, language, cache version, and evaluation policy still match; changes invalidate reuse rather than automatically deleting historical entries. TypeSafe states that it does not train on customer requests or responses; zero data retention is not the default and requires an enterprise arrangement. Its data-processing terms apply. OpenAI API data is not used for training by default; Decisions API abuse-monitoring logs may retain content for up to 30 days unless the account has approved retention controls.

Third-party services

BriefFeed's hosted backend may call a configured AI model provider to generate summaries. Those providers process the submitted payload only to provide the requested summary response for BriefFeed. If you enable Notion sync, the extension sends saved-bookmark data directly from your browser to Notion's API so it can create or update pages in the data source you configured; you can also trigger an AI Insights brief export from the side panel. For an explicitly saved Xiaohongshu note, BriefFeed may download the note's visible image files without credentials and upload them directly to the user's configured Notion workspace so Notion can render the media; these image files are not sent to BriefFeed servers. If you enable experimental NotebookLM sync, the extension can send saved items after you confirm the notebook, and can send user-triggered Insights briefs directly to NotebookLM as text sources in the notebook you configured. These external-sync payloads can include the generated summary, tags, source link, author label, timestamp, visible source text, and visible media URLs or link metadata for the saved item, plus the Insights brief text and cited source snippets you choose to save.

Chrome Web Store Limited Use

BriefFeed's use and transfer of information received from Chrome extension users complies with the Chrome Web Store User Data Policy, including the Limited Use requirements. BriefFeed uses user data only to provide or improve the disclosed reading-layer features: inline summaries, AI Insights, bookmarks, and user-enabled sync. BriefFeed does not sell user data, use it for advertising, or use it to determine creditworthiness or lending eligibility.

Changes to this policy

Material privacy changes will be reflected on this page before or alongside the relevant product change.

Contact

For privacy questions, product questions, or data handling concerns, contact hello@brieffeed.app.