Launch HN: Zenfetch (YC W23) – Turn notes/browser history into an AI assistant

136 points by rex123 ↗ HN
Hey everyone! Akash and Gabe here from Zenfetch (https://www.zenfetch.com) - a Chrome extension that transforms your saved web browsing into a personal chatGPT and search engine. Here’s a demo video: https://www.loom.com/share/abe9d40dc82545a482be55b631f9bf2a?...

How it works: (1) Optionally import your existing saved web browsing through one of our integrations or through a URL CSV import. (2) Click the Zenfetch icon in a tab to save any PDF, YouTube video, note, article, email, forum post, etc. Zenfetch can save almost anything across the internet. (3) Zenfetch indexes the text or transcript and adds that information to your personal knowledge assistant. This is done on your computer so you don’t have to worry about paywalls, logins, etc. (4) Use the dashboard or side panel to search your knowledge or chat with your personal knowledge assistant.

We built Zenfetch to solve our personal problem: we were reading tons of content but using little to none of the information. While searching for solutions, we found that most tools were good at storing information but not at retrieving it. We even found that search functionalities on existing read-later tools got worse the more content we saved.

A few examples of how our users use Zenfetch: “What was that article I read on the new nuclear fusion company?”, “Analyze this strategy based on the Lenny Rachitsky article I read”, “Summarize Karpathys video introducing LLMs”, “Which research papers mentioned the increase in carbon emissions?”, “Compile the different perspectives I’ve read on climate change”, and more.

Zenfetch is free for the first 14 days and then costs $14.99/mo (No credit card needed for the trial).

We’d love for you to try it out and let us know what we can do to improve your experience! While we only work on Chromium-based browsers right now, we’re actively working on browser compatibility and integrations. Let us know which ones to prioritize!

98 comments

[ 3.2 ms ] story [ 168 ms ] thread
'GPT assistant' is now associated with OpenAI's GPTs for me. Very confusing.
That's fair. We've considered a few other terms such as "personal AI," "browsing copilot," "personal knowledge assistant." Do any of these feel more intuitive?
Edit: changed to say AI now
OpenAI just got denied on their Trademark for GPT

There are many other GPTs from outside OpenAI, such that GPT is the contemporary Escalator

How does this compare to something like Rem [1] [2]?

[1] https://news.ycombinator.com/item?id=38787892

[2] https://github.com/jasonjmcghee/rem

Hey, good question. Rem and it's closed source competitor Rewind capture everything on your screen. Zenfetch only captures information that you explicitly add to Zenfetch. We've found that using a curated approach as opposed to capturing everything allows us to do two things: 1. Return much higher quality answers than Rem/Rewind as the information we store are items you thought were important at some point. 2. Allow us to help you remember that information with proactive features like our recap emails.
I have been building something similar [1] with the goal to make it easy to save and share things you find. I will not use SaaS for saving content anymore as I find myself wanting to build larger compositions of work and dealing with APIs just isn't it. Maybe the SaaS works today, but you will always end up wanting a plaintext, offline-first experience.

[1] https://github.com/justshare-io/justshare

This looks cool! A few notes on our approach to Zenfetch:

1. We're not trying to become your personal knowledge management solution, we want to be the layer on top that helps you retrieve and use the knowledge you've stored.

2. The biggest reason we don't do this locally is we've found hosted models are much more powerful for the knowledge retrieval and synthesis use cases. However. we do hope to move more local over time.

Depending on who your market is (sales leads, researchers, developers) it may not make sense to support local/OSS. Posting this on hn will always bring a bearded dev, me, into the comments to rant about the way things should be, but this doesn't always reflect the path to product success. Right now your tool is too general, find a use case and crush it. Find some inspiration from this ext: https://chrome.google.com/webstore/detail/workona-spaces-tab...
Thanks for the thoughts. I actually use Workona and love it haha
does this have ml to search data?
Very cool! Index my Google Email, Drive, and Calendar. "Find the receipt for that TV I bought from Best Buy 7 years ago."
Thanks! Right now don't have automatic integrations for those yet but we do allow you to individually save emails and docs. Full integrations and sync for Drive and Email coming soon!
So at present one needs to export a Google doc as PDF first? But then it won't have a stable URL, seems to me. Is there a way to upload (e.g. drag and drop -or- using a local folder) a PDF directly?
Yep we do support PDF upload today
No need to export the Google doc as a PDF. Zenfetch can save Google docs directly by clicking the zenfetch button once your on that page :)
I used to be the guy who would save cool websites I stumble across so I could 'read them later'. Lol, ya right.

Fast forward to today, I've been using Zenfetch for a few months now. Basically, everything I read, or wish I had the time to read / watch (they support Youtube too) gets saved on my Zenfetch library.

Being able to ask for summaries of stuff you are browsing in real time is great for picking up new knowledge.

Chatting with your entire library with 1,500+ saved articles and drawing correlations between different subjects + never forgetting anything, is next level.

Don't be the 'bookmarking this for later' guy.

Reading is marinating. Summary is a taster. A lot of people conjoin the two but eventually people will get smarter and understand they are not the same.
Interesting. Just for context Autonomy (now part of HP) built this cool app called Kenjin back in the 90s it used Bayesian inference applied to your browsing session to suggest content.
That's awesome! We haven't explored content discovery too much right now, since we're more focused on re-discovery of content you've already saved.

As a statistician by training, I appreciate how the traditional inference techniques were helpful way before LLMs became mainstream for these tasks :)

When will this be available on safari?
It's part of the roadmap, likely releasing in Q2

Is that your browser of choice?

Safari is my current everyday browser, but anything that isn't based on almost anything modern that isn't tainted by the chrome should be fine.

Just like IE 6 I refuse to touch it if I can at all prevent it.

(Feel free to join in folks. We can do this, IE was much better entrenched at its worst.)

Understood. We should have specified that Zenfetch works on chromium based browsers which include Chrome, Microsoft Edge, Arc, Opera, etc.

Hear you on the safari request and we will definitely expedite that release

Would love to try this (have a less sophisticated pre-LLM approach running locally), and also use Safari as my main browser. Instant paying customer for Safari, right here.

Also find myself having a pretty severe allergic reaction to websites (Gem is a recent example) roadblocking anything but Chrome (seemingly both because of their web app literally not functioning properly in Safari, and their browser extension being Chrome-only):

a) essentially forcing me to use Chrome for one task, whilst deal with password management in two places, etc., and

b) giving me the nagging worry that the design of the app or extension might contravene some non-technical standard imposed by Apple which Google does not honor…

Completely understand and we definitely want to support more browsers including Safari.

The process for exporting chromium based extensions to safari is not super straightforward (HN Community please prove me wrong!), though glad to hear it's a priority

its the default browser on my phone especially when i open links from other apps, if that is not captured, then its not much use for me
Good to know. Sounds like you primarily want a way to save articles from your phone then?

If mobile saving is what you're interested in, we can likely get that shipped even faster :)

It’s interesting.

How do you see this comparing to Pocket, Readwise, and Dewey? I think many Read Later apps will add AI search on top

Great question. A couple ways we see ourselves differentiating: 1. We're adding more and more integrations to become your knowledge assistant over your entire knowledge not just what you read online. This includes your notes, content, emails, etc. 2. We're working on doing more than just search over your knowledge. Many of our users use us to brainstorm topics for new content, compare and contrast perspectives across articles, or even create first drafts of content. 3. We're implementing features soon that will allow our model to create a "preference graph" for each user so that overtime the responses get more and more personalized to what you're looking for.
Readwise's Reader is the best annotation tool, they integrated the great highlighting UX of Reflect, but also did a decent job on how you then manage what you highlight. I don't want to click two buttons to ingest an url into both Zenfetch and Readwise, are you planning an integration / open API?
I actually use Readwise as well! For a workaround I imported all the URLs from Readwise into Zenfetch but a full Readwise integration is on the roadmap.
And can we opt out of the extension injecting things into pages? I don't want your summary, Kagi does it for me without tainting the page, I don't want to chat about this page, I don't want anything but to send this page into kind of a bookmark place, and you to scrape it and give me a smart search.
Yes! In the dashboard (https://dashboard.zenfetch.com) you should see a settings icon on the bottom left. There you can disable the zenfetch button from appearing on the page.

Then you can click on the zenfetch extension popup to still save a tab :D

I’m using zenfetch intensely for research using my own work (hundreds of podcast episodes and blog/forum posts) and others’ (books, articles, and more). Results so far are largely excellent, with the occasional bug fixed quickly. RAG as a service is tremendously beneficial and zenfetch means you don’t have to roll your own.
Can I manually add notes too? I have been looking for something exactly like this.
Yep you can! We support adding notes form Google Docs and Notion. You can also upload PDFs. Are your notes in a different format?
There's a nice tool called Glasp
Thanks for mentioning Glasp :)
Looks great! At first glance, this appears to be the service I've been searching for. I'm currently using raindrop to bookmark items I intend to read later, but its search and recollection functionality leaves much to be desired.

I'm definitely going to give it a try. However, for it to become a permanent part of my routine, it would need a Safari browser extension and an iOS app (compatible with both iPhone and iPad) to allow easy sharing and saving of content from within other apps. Stuff that Raindrop supports.

Additionally, is there an API available so that I can manually script or route content from currently unsupported sources?

Thanks for the feedback! We're working on making our extension cross-browser compatible. For the iOS experience, what's the highest priority for you: being able to save things from iOS or access chat/other functionality?

No API just yet, but if there's an integration you're looking for let us know and we will prioritize that. Thanks!

For iOS it's being able to save things first. The ability to use the chat on iOS is a second priority for me.
Perfect we’re working on a way to save from mobile that should be ready by the middle of next week
This is pretty awesome. Memex world
Thank you! Vannevar Bush is a huge inspiration for us
This reminds of Rewind [1]. I'm not affiliated to them in anyway, but I have been using their product for a few months and finding it quite useful. Their approach of screenshotting everything and saving it locally is more convenient for me, at least.

[1] https://www.rewind.ai/

Thanks, we actually used Rewind in the past.

We simply found that there were things we didn't want in our first brain, let alone in our second brain. That's why we've taken a curated content approach.

Rewind captures everything, while Zenfetch only stores the content you've explicitly saved to signal there is value.

I don't need to be reminded of the accidental clickbait article I've opened :)

doesn’t that run locally as well?
I think the data capture, compression and storage is, but the AI part isn't. If I understood it correctly, the latter is done by sending data to ChatGPT.
I’ve been using this for a few weeks. I converted to a paying customer. I’ve imported some but not yet all of my bookmarks.

I’ve been collecting bookmarks in Evernote and now obsidian for about a decade. I try to add as many tags as I hope will allow me to find an article again later. It’s often many months before I need to find something. My success rate at remembering a term from the title or the right tag is not great. I’ve been pretty impressed with zenfetch’s ability to search and find exactly what I was looking for. And this doesn’t even scratch the surface of what it can do when you want it to synthesize answers from many articles for you.

They’re also working on indexing your saved tweets which I’m excited about. It’s a giant pain to try to find liked tweets.

Thanks! Saved tweets is our next integration : )
Does this have the hallucination problems that LLMs show? Do you offer any protection against that?
We do use the standard hallucination checks to make sure the final generated answer is rooted in the retrieved context.

It is an ongoing effort to ensure truly factual answers and this is one of our top priorities :)

For my primary usage, suggesting topics and ideas based on what I've written and read, hallucinations are harmless -- if zenfetch suggests a silly or baseless idea, I just ignore it look at the other nine. So yes, but it doesn't really matter for me.
I am pretty sure this is what Microsoft is trying to do.
Totally fair. Our priority is allowing you to index anything regardless of what ecosystem it came from
They've been trying to make software for a long time.
Perhaps I’m not understanding how this stores data, but how do I retrieve source material after I give it to the app?

Eg. Let’s say I adopt Zenfetch, and start using it as the sole place where I keep bookmarks. Then, something happens such that I want to retrieve that original, principal, bookmark data.

How do I go about retrieving/exporting that?

We will enable an export data option soon where you can download a list of the bookmarks you saved to Zenfetch.

Our goal is not to become a vendor lock-in play. Instead, Zenfetch is a layer on top of your knowledge base goaled on helping you activate information that you forgot about.

If you're referring to being able to revisit bookmarks: from the dashboard, you can click on any of the articles to see the original article

Let me know if more clarity is needed

just a list of urls or the actual content?
Zenfetch will capture the actual content of those URLs (assuming they still exist and are accessible from your browser)
Not the original poster, but still not clear to me.

Say I mark a page that contains really usable information. For some reason, 2 months later that page goes offline because the domain was not renewed (or whatever). Some time after that happened, I need that information. Will I still be able to see the original content, or not?

Similar thing, I mark a page. The author updates the url with different content 2 months later. Some time after that happened, I need that information. Will I still be able to see the original content, or will I just find the new updated content?

I guess the question is: is a snapshot taken from the actual content at the time I mark it as interesting and can that snapshot be shown, or not?

Snapshot is taken from the actual content at that point in time. We haven’t enabled a reader format just yet where you could view the original text. Right now, clicking the card in the dashboard will redirect you to the URL.

If you were to chat with the article from the dashboard, it would be preserving the snapshotted content and leverage that information in the final answer (same with using the search functionality).

Clear, thanks!

I did not test it yet, but it looks like something that would be very useful for me. So, I will definitely try it.

I will add that me too, I have some concerns regarding the data being "locked" in your service, but I read in other comments that at some point it will be possible to create an export of some sort. When using this to carefully craft a knowledge base during months, it would be very painful if that is all lost when stopping the service for whatever reason.

As others asked, I would appreciate an Android/iOS app to register new content when reading something on the phone.

Regarding the pricing, it mentions a fixed price per month, but is there any limit on the amount of data that is uploaded (for example when uploading PDFs) ? Is there any limit on the amount of content in the knowledge base?

Appreciate that and yes in the absolute worst case, you can email us and we can manually export your data on your behalf.

We are using a fixed price today and have no restriction on storage. This might change in the future if costs scale, though it’s not an immediate priority and we’d be sure to communicate those changes well in advance.

Once you’re onboarded, feel free to message us with the in app support widget. It’s a direct line to the team slack and we tend to respond almost immediately

I am trying to create an account (using a gmail address as email address), but I do not receive a code through email. I already tried multiple times. Any idea what is going on?
Sorry about that, this happens sparingly with our auth provider.

It might take a couple of minutes. Would recommend checking spam or trying again in a bit

Hey Akash & Gade, Whom do you think is the ideal ICP for Zenfetch?
Hey great question. We consider the following a good fit for us: - In terms of role, we’ve found prosumers (specifically content creators) get a lot of value out of being able to take their previous writing/research and generate new ideas or rough drafts - People whose job relies on collecting a lot of information such as founders, consultants, coaches, etc. have also been power users - Finally anyone who tends to consume a lot of content whether it’s videos, blogs, newsletters, or books has found value in being able to quickly find the source or context for the idea they were thinking of
As I split 80/20 between mobile and desktop browsing; any ideas on a mobile version?
We’re actively working on a way to save from mobile but chat from mobile will take a bit longer
Are there any plans to releasing a desktop app? It would be nice to combine the current use case with a "chat with my documents/pdfs" flavor as well.
Gabe/Akash will tell you their plans I'm sure, but I'm already using it to import PDFs and other documents via the web, so I don't feel the need for anything on the desktop. (A local version would be stupendous for client confidentiality, but I'm not holding my breath for that.)
Couldn’t have said it better myself! We do want to support a local version in the future as local models get better. However today you can upload PDFs. If there’s any other file types you want to upload let us know!
Safari support would be great. Any chance of a self hosted version?
I second the self hosted version if possible, perhaps with a 7B model! My main concerns are privacy and control.
We're actively thinking through ways to support a local version. Most of our users prefer being able to access the best models but seems like Mistral will have something comparable soon.
Could it work with bookmarklet as like what pinboard did?

Also any support for importing from pinboard archives?

Haven’t considered bookmarklet, what would that make better for you? Also you can import a CSV of URLs today if Pinboard has an export option