Hidden ramen spots in Tokyo you must try
@tastingtokyo
AI Summary
Six under-the-radar ramen shops across Shinjuku and Shibuya, with the dish to order at each and how to actually find them.
#tokyo#food#travel
Food & TravelMar 12
Ask your library
Ask your library
instead of scrolling it.
"What did I save about Tokyo?" is a better interface than a grid. Nemivo retrieves the saves that actually answer the question, reads them, and writes back an answer that cites each one.
Start Free 300 saves free · No credit card required
In short
Ask is a question box over your own saves. Nemivo retrieves the handful of saves that bear on the question, gives only those few to a language model, and streams back an answer with a numbered citation on every claim. It never searches the web, so when the answer is not in what you saved, it says so instead of inventing one.
what was that video about pricing psychology?
Nemivo
You saved “Why $9.99 works” three months ago. A 12-minute talk on anchoring and charm pricing. Your note says “use for the launch page”.
Why $9.99 works: the psychology of pricingThread: 12 pricing-page patterns that convert
Retrieved first,
written second.
The order matters. A model that writes first and looks for sources afterwards is a model that invents them.
- 01
Your question becomes a query
The intent is pulled out first: what you are asking about, and any period you mentioned. Relative dates like "last month" are resolved locally rather than taken from the model, so the range is never invented.
- 02
The library is retrieved, not read whole
The same hybrid retrieval that powers search selects the top sources: four on fast, eight on balanced, sixteen on deep. Your whole library is never sent anywhere, only those few saves.
- 03
The answer streams, with citations
The model is given those sources and told to answer from them alone, citing each inline. The answer arrives token by token, so a deep question reads as it is being written.
- 04
Keep it, or ask the next thing
Questions live in a conversation you can come back to, and any answer can be saved as a note with the sources it cited attached to it.
The workspace
A research desk, not a chat box.
/ask is a page of its own in the app, built around the fact that the interesting questions take more than one turn.
A citation on every claim
Answers cite the saves they used as numbered references, and each one links back to the item so you can check it in one click.
Three depths
Fast reads four sources, balanced eight, deep sixteen. More sources is a better answer and a slower one, so the choice is yours per question.
Conversations that persist
Threads are kept, so the follow-up question three days later still has the context of the first one.
Answers become notes
Save an answer and it lands in Notes with the ids of the saves it cited, so the reasoning survives the conversation.
Grounded on dates
Each source carries the date it was saved, and time-scoped questions are told the retrieved set is the complete set for that period, so the answer lists exactly those and does not imply others.
Your library only
There is no web search behind this. If the answer is not in what you saved, the honest answer is that it is not there.
The limits
What it cannot answer.
A retrieval-grounded answer is only ever as good as what it retrieved, and pretending otherwise is how these things go wrong.
It never leaves your library
No web search, no outside sources, no general knowledge dressed up as your own. That is the point of it, and it is also the limit: a question about something you never saved has no answer here.
It reads the summaries, not the video
The sources handed to the model are what the pipeline already holds: titles, creators, tags, summaries, dates and links. A save the AI could not read well makes a thin source.
Two credits a question
Asking costs two AI credits, because it is a retrieval plus a model call. Free gets 100 welcome credits and 20 a month; Pro is unlimited.
Notes are not retrieved yet
An answer is built from the saves, not from the notes you wrote on them. Notes are searchable on their own page, and folding them into the retrieval is not something we have done yet.
The long version, including what is stored and where the AI sits, is on how it works.
It only works because the saves were read.
Every source in an answer is a summary the enrichment step wrote and a vector the embedding step stored. Ask is what those two steps were for.
Pricing
Start free. Stay for the memory.
A real library on the free plan, forever. Not a trial: no countdown, no card, and no email on day seven asking for one. Pro removes the limits when your memory outgrows them.
Free
$0forever
A plan, not a trial. It does not expire.
- 20 AI credits a month
- 300 saved items
- 10 collections
- 1 automation
- 10 AI searches a day
- JSON export
No card. Ever.
Pro
Recommended$7.99/month
Built for people who save content every day.
- Unlimited AI organization
- Unlimited saved content
- Unlimited AI search
- Unlimited automations
Everything in Free, plus
- Faster AI processing
- AI insights in exports
- AI automation conditions
- Export as JSON, CSV, Markdown
- Public share links
- Bulk library reprocess
Cancel anytime
Free includes 100 welcome AI credits · Secure checkout by Polar
FAQ
Ask your library questions.
No. Your library is never sent anywhere as a whole. A question retrieves a handful of saves, and only those few are given to the model, with an instruction to answer from them alone and cite them.
Want the mechanism and the limits in full? Read how it works.
Remember everything.
Forget nothing.
Nemivo reads the caption on everything you save, writes the summary, adds the tags and files it away. Months later you ask for it in your own words.
Start Free
A free plan, not a free trial · No credit card · Nothing expires
