The Fall of Roam (2022)
every.to
every.to
I think this might be a remnant of the time we did not delete media for graphs for cases we thought they might've just migrated to a new graph. For context, a semi-common pattern was for users to export their graph and restore to a new graph, so that they can change the name. Could you have gone through a similar process before deleting your account?
If you please contact support@roamresearch.com and provide the firebase links (even just a few should be okay to find the media), then we can proceed with the deletion for you. Sorry for the issue
I work at Roam on the engineering team.
I do not claim to know about this case, could you send me or support@roamresearch.com any more details you have re: this?
I can, however, tell you what the protocol has been since I've been working here at Roam (since 2021). No one can access user notes without an explicit written permission being granted. We have logs for when any graph is accessed via admins, and so, any member on the team accessing user notes without permission would be fired immediately. This was the operating policy and was made clear to me on my onboarding itself, along with the policy of immediate termination in the case of abuse.
Additionally, since Jan 2022, we have the ability for users to create End-to-end encrypted graphs. These graphs provide an extra level of protection - where your notes (& media) would be safe even in the worst case of Roam being hacked or compelled by law agencies to give info (to be clear, we haven't had either happen)
Roam actually DID NOT READ THEIR DATA (we have always had the policy of never accessing user data without explicit user permission). She just misunderstood what she was looking at.
More (verifiable) details in my comment here: https://news.ycombinator.com/item?id=44047945
I think I found your account and I don't see any access logs to your graph from anyone other than your account. If you can provide any more info or screenshots of we would be able to dig deeper into exactly what you saw. It could have been a console log or a hard coded employee email in the code.
We've always cared deeply about user's privacy and ownership over their notes. This is why we've had this policy from the start and focused heavily on local first features and data portability. We offer fully offline graphs, where the data never touches our server and is never able to be accessed by anyone on our team. We also offer fully encrypted graphs, which are stored on our servers but are not able to be read by anyone without the password (our team cannot read your data).
Emails: https://www.dropbox.com/scl/fi/s6ed1brrcvc0hncig7nm0/IMG_205...
https://www.dropbox.com/scl/fi/ohafavhr9nlqfedlbfxrd/IMG_206...
We followed up with our ex-employee to get the final (cutoff) message in this email thread https://firebasestorage.googleapis.com/v0/b/firescript-577a2...
In it he tries to explain that these are help graph transactions they are seeing. I do apologize if you didn't understand it at the time, but we did try to explain it to you. I Hope this clears up everything for anyone following along.
Roam actually DID NOT READ THEIR DATA (we have always had the policy of never accessing user data without explicit user permission). She just misunderstood what she was looking at.
More (verifiable) details in my comment here: https://news.ycombinator.com/item?id=44047945
But in the spirit of constructive criticism, your user had been carrying that belief for 5 years now because no one explained it when they originally reported the finding.
I understand what the screenshots are saying and this makes it clear that it was a misunderstanding and that NO ONE ACCESSED YOUR GRAPH(S). Please let me explain
Lets start with your first screenshot: https://www.dropbox.com/scl/fi/g9jv8eh1ugi5qda0c6azx/0811202...
If you take a look at this screenshot, it shows that the values you saw are in the indexeddb db "..._help-tx". The "help" bit denotes that those are the actions/txs taken in the "help" graph (which you can access via https://roamresearch.com/#/app/help). The reason you're seeing Bardia and Conor's emails there is because they wrote in the help graph (maybe they were writing guides there or adding stuff to the changelog). The reason the help graph data is in your indexedDB is because you probably opened the help graph at some point.
If someone had accessed your graphs, similar txs would have shown instead in the indexeddb dbs "..._DZ-tx" or "..._programming-with-categories-tx"
Everything I've said above can be verified if you say go to any Roam graph, and see what dbs are stored in IndexedDB in the devtools.
Hopefully this makes sense. Also, as Bardia replied in the email, we have never and will never edit user notes without explicit permission.
tl;dr: You thought you were looking at the logs for your graph but you were looking at the logs for the "help" graph. This is easily verifiable from your screenshots itself if you know where to look (details above).
What is our fault that we did not clear up this misunderstanding immediately and I'm sorry for that dzink (I think this was during the time of Roam hypergrowth so maybe Bardia missed the later email reply?)
Hopefully, it is now clear that NO ONE accessed your graphs (and no graph is ever accessed without explicit user permission)
P.S. I took this video & wrote this message alongside my earlier one but could not post because HN said "You're posting too fast. Please slow down" XD. Hence the new account. Please do not block me because of this, mod
Don’t do that. Things like this are an opportunity to overcommunicate. Explain in detail to the user, who’s smart enough to use the tools to reveal the information in the first place. Write up a FAQ entry explaining this for the next person so you can point them right to it. Don’t just reply to say basically “you didn’t actually see that” and leave it at that.
Yes, definitely a communication misstep on our part, and we could've handled it much better
I have a second comment there too, but got delayed in posting because HN rate-limited my account for too many comments
https://firebasestorage.googleapis.com/v0/b/firescript-577a2...
(looks like this will be my last reply in this thread)
This can wait until tomorrow.
Roam actually DID NOT READ THEIR DATA (we have always had the policy of never accessing user data without explicit user permission). She just misunderstood what she was looking at.
More (verifiable) details in my comment here: https://news.ycombinator.com/item?id=44047945
I've been an active user for a couple of years now and have substantial amount of information stored in Roam. I guess I should have known better than to have sensitive data stored in someone else's servers without encryption.
Time to explore Obsidian and see what the migration path looks like.
I want to clarify this (since you might not see the other replies in this thread)
Roam actually DID NOT READ THEIR DATA (we have always had the policy of never accessing user data without explicit user permission). She just misunderstood what she was looking at.
More (verifiable) details in my comment here: https://news.ycombinator.com/item?id=44047945
Have they clamped down on employee access? Was this "new employee" let go for accessing user data without any apparent reason?
Relevant reply here: https://news.ycombinator.com/item?id=44038085
Roam actually DID NOT READ THEIR DATA (we have always had the policy of never accessing user data without explicit user permission). She just misunderstood what she was looking at.
More (verifiable) details in my comment here: https://news.ycombinator.com/item?id=44047945
- Just start typing
- Let the LLM analyze what you're typing, given the RAG database of everything else you've added, and be able to make those kinds of correlations quickly.
- One-button approve the backlinks that it's suggesting (or even go Cursor-style yolo mode for your backlinks).
Then, have a periodic process do some kind of directed analysis; are you keeping a journal, and want to make sure that you're writing enough in your journal? Are you talking about the same subjects over and over again? Should you mix things up? Things like that would be perfect for an LLM to make suggestions about. I don't know if Roam is thinking of doing this or not.
You're thinking of an optional step of adding extra links "just because", but IMO that's as a learning process in the beginning when you're not used to adding any forward-links whatsoever.
IMO the 3 table-stakes features for a notetaking app in 2025 are AI-powered search (including a question-answering capability), showing related / recommended notes (via RAG), and automated clustering (K Means + LLM) to maintain a category hierarchy.
Can Ollama do this yet?
I had a similar problem when designing my personal management system last decade [1]. Every system you use, you have to stick to in order to get results. Sticking to a system can be emotionally draining to the point where you give up.
IMHO, that sense of emotional drain you get with fancy note-taking systems is tapping into something true. Only a small fraction of what we think we need to remember actually matters and will benefit from so much care to structuring it. The rest is a waste and a drain on our limited cognitive resources.
My solution is to initially write in a designated place that allows for less structure. In the to-do system, the main doc has a "landing zone" for action items to be quickly jotted down, then structured and organized later. In the project system, I'll have a "dump" file where I dump project thoughts that I'm not sure are important. I trust that if the ideas I jot down are actually important, the structure they deserve will come to me later.
Is that trust always right? Maybe not 100% of the time, but it seems like a more useful heuristic than "everything I put into this system needs lots of structure I don't feel like providing, so I don't, and it makes me feel like a failure".
[1] https://open.substack.com/pub/renormalize/p/my-markdown-proj...
I don't know about you guys, but I'm an Obsidian lover and that's not gonna stop anytime soon. IMO the big problem about what this guy is saying can be boiled down to this:
>My most common behavior is to Actually Write the Notes. That’s why Roam needs to help me with the thought, ‘I don’t know where to put this.’ If it does that well, it makes the vast majority of my time spent in the app a breeze. If it does that poorly, it makes my experience so painful that I want to switch systems.
The lesson of Obsidian for me has been that organization is creativity. If what you want is to have an ideological maid that can organize all your thoughts for you, then you're gonna have a bad time with any note taking service (although I'm sure you can develop llm plugins to do this in a way that you personally enjoy now.) What's beneficial about these note apps is that they put this issue directly in front of your face. Either rise to meet it or go back to pretending like organization doesn't matter and avoid the responsibility of creativity.
Using Obsidian goes through stages much like a growing business. You start and you have personal relationships with all the notes so you can remember them, but once you get enough notes you realize it's too much to manage just using personal relationships and you need to start implementing a system. As you get better, your system changes, leaving a paper trail of notes with different systems. That's why the only thing that I think these note apps need is a deprecation system, but otherwise IMO they're perfect.
Steph Ango, CEO of Obsidian, has a nice article on how he uses Obsidian.[3]
1. https://brajeshwar.com/2025/obsidian/
You can use folders, tags, properties, links between notes (exporable through the links panel per file or the graph view), and there are extensions that let you add more advanced functionality. In the end, any system will require you to come up with your own system of organization.
Roam bet on the idea that if you link enough atomic notes, structure will self-organize.
Which is such a weird fantasy if you spend a few minutes thinking about it. Try writing code like that or building a company or just about anything else! Why should notetaking and archive development be any different
It's clear you need some sort of editorial hand to create something maintainable and future proof. Like zettelkasten had Luhmann’s obsessive discipline behind it. Evidently roam had um. enthusiasm and javascript?
and yeah, it’s telling that the comparison is to IDEs. Imagine an IDE that dumped every snippet you typed into a graph database and expected you to recompile coherence out of it by browsing links. thats what roam felt like after the honeymoon.
In general most of Roam's target should want to lean harder into opinionated workflows. there’s a reason tools like linear or notion are winning. they’re structured enough to relieve cognitive load, flexible enough to adapt. Roam tried to be emacs, but turns out most users don’t want to configure their own productivity dialect.
also, lol at the idea of "automated taxonomy". The entire knowledge management industry keeps rediscovering ontologies like they’re new. We are probably going to reinvent OWL at some point and give it a name like "neuroschema" or something
Roam implemented static bidirectional links and called it associative memory. in reality, it's closer to mind-mapping software with backlinks. So without mechanisms for reinforcement (surfacing old notes intelligently), pruning (forgetting irrelevant junk), or plasticity (reorganizing in response to use), the system becomes a junkyard of half-formed thoughts.
Brains forget for a reason, roam doesn't
I think this is the key mistake in Roam's design (and in many ways, obsidian and friends). They appeal to a dream some people have that maybe if you never forget anything, you'll get smarter forever. (Or something like that).
The problem is that there's many benefits to having a mind which forgets things. That property lets us grow and change over time - and move on from old ideas or old ways of thinking. Not necessarily because they're bad; but because we become a different person from the person who had that thought.
Trauma is an extreme case of this. Its essentially a disorder of memory; where we etch some old memory in stone. Because we don't let ourselves forget it, we inevitably build structure / thought patterns around that memory. "This one time __" - "As a result, deep down I believe that I am fundamentally ___ (unsafe / unworthy / stupid / unlovable / ...)". Trauma work is in many ways a slow process of learning to unclench your mind from those past experiences, to allow yourself to "move on" from them. (Ie, forget the emotional impact they have today.)
Its also kind of obvious in software or architecture. You can't just keep adding to an old structure forever. Software gets harder to build the bigger it gets. Same with buildings, books, teams and more. If everything new needs to fit with everything that has come before, its an O(n^2) job. Of course roam suffers from this too. The default "remember everything forever" default is naive and silly. Our brains don't work best like that.
You are just making a very silly "Appeal to nature" argument. Your notes, just as your memories, change and morph. For your memories, every time you access them, for your notes, every time you notice something you could improve. Old notes should not bother you, just ignore them if they're not relevant. They take a negligible amount of space on your devices. Personally, every note I've taken serves a purpose, even if their purpose is to just fill a spot so that I may be continually aware I've tackled a particular subject before even if it has not had any relevance for years.
I don't see it that way. I see it as a healthy, useful expression of continuous death.
In software, we don't start every program by first importing every line of code ever written. Why not? The computer has room for all that code. Why don't we import it all into our workspace? The reason, in my mind, is that each line of code in a computer program has a cognitive cost to it. A sort of, conceptual gravity, which makes reaching for further away ideas much more difficult.
When brainstorming, often a blank page is the best canvas for a new idea. We start companies with new stationary. New workbooks. We even have sayings for this - "Blue sky thinking" or "Greenfield projects". Ie, projects which don't inherit older, more established structures or code.
There's a balance of course. We also don't start everything from scratch either. In code we pull in libraries as we need them, and lean on our programming languages and operating systems. But you have to strike the right balance between new and old. Too much old and you're stifled by it. Too much new and you're trying to boil the ocean.
I think humans are like that too. I think our ability to crystalize new thoughts depends on our capacity to let go of old ones. I don't think the best minds spend their lives hoarding all the best knowledge. For my money, the old people I like the most are people who can be in the here and now. Knowledgable, sure. But also present. Open to surprise. Philosophically you want to combine whats happening right now with the best ideas from the past. And let the rest go.
At least, that's how I think of it for myself. If I'm a different person in 20 years from who I am now, I wish whoever I become the best of luck. I hope for them to be unburdened by all the cognitive misadventure I'm probably going through right now.
That's physically impossible.
> At least for me — and most of the people I know — we got a garbage dump full of crufty links and pieces of text we hardly ever revisit. And we feel guilty and sad about it.
It'll never work if you can't leave things behind.
Really, I think the user in that case needs to be much more choosy about what they put in the database. It will save them time and greatly improve the signal-to-noise ration.
Going through all your notes regularly and asking “should I delete this? Does it bring me joy?”
I'd just delete things opportunistically, as you come across them. If you aren't coming across them, there's no value in deleting them.
This is so true. Regardless of how useful note taking actually is, the kind of people using these apps are those who like the idea of having everything "perfectly organized" - and this friction and uncertainty of where to put notes gets in the way of that. I'm the same. Every time I know that I don't have a proper place for a note I stop taking notes alltogether. I guess that's for the better.
Perhaps what he needs is for the tool to automatically ask him "Is it okay to delete this note from 60 days ago?" That should be long enough for him to lose any attachment to what he wrote and a lot of the time he should say yes, and delete the crap.
OTOH the app that really won was Obsidian, due to flawless execution with the "local first" principle. Being closed source and "not listening too much to the community" weren't issues, they just focused and improved consistently.
I debunked this myth on the prior discussion in 2022. Backlinks were a well-known idea in the wiki community, to the point that they were part of WikiMatrix, and it's almost certainly the case that Roam copied the idea. TiddlyWiki had backlinks at least as early as 2006. https://news.ycombinator.com/item?id=30330835
For me Roam's killer feature was transclusion. When they launched, no one else had it. I could write all my notes in the daily notes and still have a sufficiently-well-organised knowledge base for specific subjects, tasks, projects.
These days I use obsidian for the simplicity/portability of .md.
Roam also copied the idea of transclusion. I'm not sure who is included in "no one else" but Wikipedia's had it pretty much from the beginning. See this page from 2005[1]
The term was coined in a book in 1980[2]. This page from 2007[3] says "...CvWiki, developed in 1997 by Peter Merel, which was the first Wiki clone to have functioning transclusion, backlinks and WayBackMode."
[1] https://en.wikipedia.org/w/index.php?title=Wikipedia:Transcl...
[2] https://en.wikipedia.org/wiki/Transclusion#History_and_imple...
[3] https://en.wikipedia.org/w/index.php?title=History_of_wikis&...
I'm talking in the practical sense here. It's hard to take you seriously when you mention Wikipedia in the context of personal note taking apps.
Why do people tolerate the WWW working like this?
And your perspective today might differ tomorrow or in a year or 20 years. Think about Wikipedia (and other wikis) - there is no hierarchy. You can start at any point and, in a sense, there's a hierarchy of pages with the starting point at the top.
The Dewey decimal system has less ambiguity in both, and an alphabetical system would be unambiguous for archiving (if not retrieval).
I prefer to organize my notes functionally (eg internal emails, blog posts, links to read, reading notes) and then rely on search for retrieval. It’s not perfect but it lowers the friction, which I think is very important.
True, but .......
Full-text search solves many problems. And hyperlinks enable knowledge to be in multiple places at once and even remain normalized (i.e., it can be in other locations by reference). If your kb has an efficient 'include' functionality, it's even easier.
Most importantly, if you are using your kb well (by my definition [0]), you record high-quality knowledge that you've already engaged deeply with. If you can't solve the content ambiguity issue easily, it's just a sign that you haven't engaged and don't know it well enough.
Like if you want to brush up python for interviews, you're gonna want to take notes about specific things like heaps and string builders. You don't want to dilute that info with stuff you know will never be asked in an interview like how to build a TUI.
(I haven't used Roam personally and have no idea if it even has a plugin architecture or is extensible, but this reminds me a lot of some of the knowledge management work we're doing with corporates)
Make topic-specific folders for discrete topics (e.g. recipes). Anything generic, put it in a big diary file with daily entries. It's easy to scroll through my past few days of notes, and after a few days I don't really need to reference uncategorized miscellanea for the most part. If I do, I can usually find it with ctrl-F-style text search.
At the end of the day/week/whatever, feel free to #tag anything you think you'll need to come back to or copy it into a topic-specific file. I mostly don't do this though. People feel a need to retain this big body of knowledge from their notes, but I think most notes are disposable. It's easier to wait a little while before reviewing & then decide what's worth saving, which is typically not much.
I ended up building my own app for my notes and it turned into a chronological feed of short notes, like a social media feed. I just recently added linking between notes, but honestly, I've found that it's not essential. Just having a way to search by text and tag covers most of my needs. The chronological order also makes it easy to find stuff that I wrote recently or to filter by date.
Since there are no files or folders, there's also zero friction when it comes to recording something. I don't need to think "Where should this go?" or "Is there already a folder or a larger note this should be a part of?" I think that has honestly led me to just down more thoughts and ideas than if I was trying to maintain a strict structure to everything. (There are downsides to that, though, as it may mean I have more noise in my system, making it harder to find actual notes of value long term.)
Another one for the “not really using Roam any more” box
Roam also took absurdly long to finish loading, long enough to lose my train of thought and get angry at Roam instead of writing down what I wanted to write down. I would write it in Sublime Text while Roam loaded, then paste it in. Then wonder why I was bothering with Roam when I could’ve already saved a text file with my note, to a directory somewhere.
Do you think Roam’s team fixed the performance problems? Ha.
IMHO you may need to produce much better quality knowledge. You are missing out.
A good KB - personal or shared - captures high-value knowledge and lets you pickup where you left off, years later, with little effort. That way you are always working with the best knowledge you've ever had. What defines high-value?
First, it's high-impact - it changes things in significant ways: Trivia about C++'s origins is unlikely to be worthwhile; something from an expert that changes your whole perspective about C++'s design and applications may be. Also it's accurate, high-quality knowledge; otherwise the impact will be much reduced or it may even be significant in the wrong way - for example, Wikipedia IMHO doesn't qualify (in other ways too), but a lit review by an expert can be priceless.
Second, it's hard to replace: 1) Discovery is unlikely: you are unlikely to think of or encounter it next time, at least not unless you revisit the issue in depth. 2) It's hard to find - even if you think of it, you won't be able to find it or recreate it easily. Maybe it's buried in a book you won't remember. For example, if you have info on operating systems, you need little about Windows, Android, etc. because you use them daily (hypothetically) and info is easily available. Insights on TempleOS might be better, or from that keynote by Dennis Ritchie that you attended.
By capturing the value, you get much greater ROI - a lifetime or career of it - from your knowledge work. That also incentivizes deep, high-value knowledge work.