Recall is for people who save articles, videos, podcasts, and PDFs faster than they can use them.
It turns each source into a summary card, adds tags, links related material, and lets the user ask questions across the whole library. It can also read summaries aloud and turn them into memory quizzes.
That can be genuinely useful. It can also create a very organized collection of shortened material that nobody checks or returns to.
We would judge Recall by verified reuse. Save 50 things that matter, return after two weeks, and see whether the exact source is easier to recover. A summary is a map to the source, not a replacement for it.
The first job is complete capture
Recall cannot summarize text it never saved.
Build a test set that reflects the material you actually use. Include a simple article, a long report, a dynamic page, a page behind a login where permitted, a PDF, a PDF with columns, a table, a chart caption, footnotes, a YouTube video, a podcast, and personal notes.
For each source, check the beginning, middle, and end of the saved content. Look for table rows, warnings, exceptions, captions, links, speaker names, and time stamps.
A page may look complete while its table is missing. A video summary may be based on an automatic transcript that misunderstood a name. A PDF with two columns can arrive in the wrong reading order.
Partial capture and mixed text-and-table handling appear in a detailed paid-user account. That does not prove every source will fail. It identifies the first test.
If Recall saves only part of a source, the card should make that obvious. A short clean summary of an incomplete document is more dangerous than an error message.
Do not use Recall to bypass access restrictions or save material in a way the source owner does not allow. Check copyright, account, and workplace rules before uploading private or licensed content.
Check the summary against the awkward part
AI summaries tend to preserve the main topic. The useful question is whether they preserve the qualification that changes the conclusion.
Choose a source with a known answer. Mark five facts before saving it. Include one number, one date, one exception, one argument from the middle, and one detail from a table or footnote.
Compare the short and detailed summary with those facts. Check the linked time stamp in audio or video. Open the original source from the card and confirm the passage.
Then choose two sources that disagree. Ask Recall what each one claims and why. A good answer should keep the disagreement. It should not blend them into a neat statement neither author made.
Ask a question the library cannot answer. The reply should say the source is missing rather than filling the gap from general model knowledge. When web search is enabled, the answer should make clear which parts came from the saved library and which came from the internet.
Whole-library chat is one of Recall's main attractions. It can remove the work of opening ten cards. It raises the cost of one bad card too. Every important answer needs a path back to the item and passage that supports it.
Automatic links need to bring something useful back
Recall uses smart tags and a knowledge graph. The graph is a visual map of saved items and the links between them.
A large graph looks impressive. It is not a measure of learning.
Save ten sources around one topic and ten around unrelated topics. Check the automatic tags. Merge duplicates if the product allows it. Remove a wrong link. Add a personal note that explains a connection the model missed.
Then ignore the library for two weeks. Return through a new article or question. Count how often Recall brings back an older item that helps with the current work.
That useful return is the outcome. A line connecting two cards because both mention “AI” is not.
Augmented Browsing can show related saved material while the user moves around the web. Test whether the suggestions arrive at the right time without covering the page or slowing the browser. Make sure private card titles do not appear on a screen that another person can see.
The public feedback praises the graph and resurfacing. Long-term reliability at a large library size is still unclear. Test search and chat again after importing hundreds or thousands of items, not only with a neat first week.
Use the quiz only for facts worth remembering
Spaced repetition shows a fact again after a delay. The gaps become longer when recall improves. This can help with vocabulary, definitions, names, and other material that benefits from active memory.
AI can generate a confident quiz from a wrong summary. Review the question and answer before adding it to a study routine.
Choose 20 facts from the test library. Reject vague questions. Keep the source link with every answer. Try the review schedule for two weeks and see whether it asks the right thing at a sensible time.
A personal knowledge base is not required to memorize everything inside it. Use quizzes for facts that make future thinking or work easier. The library can hold the rest.
Test every language and browser you need
Recall advertises support for several languages. A user account reports mixed language and translation behavior.
Save sources in the two main languages you read. Ask questions in each language. Request a summary in the original language and another in a chosen language. Check names, quoted terms, numbers, and whether chat silently switches languages.
Use the real browser too. The Firefox extension has a small set of reviews that includes performance, sign-in, and reliability complaints. Install it in a clean browser profile. Measure startup and page load. Save 20 pages in a row. Sign out and back in. Restart the browser.
Repeat the core save route in Chrome if both browsers matter. Then use iOS or Android. Save from a share sheet, open a card, ask a question, and return to the source.
The extension should make capture easier without making every page slower. If the main browser becomes unreliable, the knowledge workflow is not convenient.
The plans separate normal and bulk use
Recall Free costs nothing and needs no card. It allows unlimited saved articles, videos, podcasts, PDFs, and personal notes. It includes ten AI summaries per month, a manual knowledge graph, and API plus MCP access.
MCP is a connection that lets a compatible AI tool reach the Recall library. Review the access given to any outside tool before using it.
Plus costs $10 per month when billed yearly. It adds unlimited summaries for typical use, automatic tags, whole-library chat, text-to-speech, automatic graph links, augmented browsing, quizzes, spaced review, multi-language support, and bulk import of up to 1,000 bookmarks or 10,000 Markdown notes.
Max costs $38 per month billed yearly. It adds a choice of frontier AI models, bulk actions on up to 100 items at once, and one-to-one onboarding from the founding team.
The page displays a 20% saving for yearly billing. Check the monthly rates in the live account before choosing. Students can request a 20% discount. A refund is available within the first 30 days by email.
“Unlimited” means typical use under a fair-use policy. Recall says it may pause, review, or change a subscription when activity greatly exceeds reasonable levels.
A public dispute describes a power user's summaries being limited. The company said automated extreme use had crossed fair use and offered a refund. That account does not tell us where every current boundary lies.
A normal personal library is different from feeding thousands of videos through an automated process. If bulk use is the reason to buy Max, describe the daily volume and automation to support first. Get the expected limit and response in writing.
Export before the library feels permanent
A useful Recall library becomes hard to leave because it holds source links, summaries, personal notes, tags, graph links, and review history.
Test export during the first month. Save a known set, add notes and links, then export it. Open the files without Recall.
Check the original title, author, source address, saved text or reference, summary, tags, dates, personal notes, links, and attachments. See which graph connections and quiz history survive.
Import some notes into another tool. The data does not need to recreate Recall's AI features perfectly. It does need to preserve the user's work and a route back to sources.
The current pricing page says saved content remains after a downgrade. Free limits apply to new AI work. Cancellation keeps paid access until the end of the billing period.
Review privacy before adding private work documents. Recall says saved data is private and is not used to train AI models. Check the current policy, subprocessors, deletion route, security controls, and whether the data needs stronger workplace rules.
Who should choose Recall?
Recall is worth testing for a researcher, writer, student, or curious person who consumes several formats and wants old material to return during new work. It is most useful when summaries start a review and chat leads back to evidence.
It is unnecessary if an existing read-later or notes system already makes sources easy to retrieve. It is risky as the only archive when capture, export, or large-library behavior cannot be proven.
We did not save a page, compare a transcript, query a library, use the graph, run a quiz, test a browser, export data, hit fair use, or contact support. The praise is meaningful but comes largely from Product Hunt and small user samples.
Our rule is to save 50 varied real items. Verify the beginning, middle, end, tables, and footnotes. Ask questions with known answers, conflicts, and no answer. Test two languages, the real browser, and the phone.
Ignore the library for two weeks. Then try to recover five exact claims without browsing cards by hand. Every useful answer should lead to the source.
Finally, export the notes, tags, links, dates, and source references. We recommend keeping Recall if it brings verified material back when needed. If it only makes saving more satisfying, it has improved the inbox and not the learning.
