I built something to replace two of my favorite apps for PKM
Everything I save lived outside my vault, which always bugged me, so I solved that by overlapping the two apps I already pay for. What I like about each: mymind: the visual grid, the minimalism, never filing anything Recall: the summaries, especially on…
Everything I save lived outside my vault, which always bugged me, so I solved that by overlapping the two apps I already pay for. What I like about each: mymind: the visual grid, the minimalism, never filing anything Recall: the summaries, especially on YouTube where I rarely ever rewatch the information, but I want to connect the ideas. I built a tool for myself where everything is stored in my Vault, lands as plain markdown, and gets summarized locally through Ollama. Where it stands: extension saves pages, images, PDFs, YouTube with the full transcript, Instagram hover any image on any site, a save button shows up a local model writes the summary and tags, and actually looks at the images search runs three passes: exact, then expanded (bw finds black and white), then semantic "more like this" on every item, from embeddings paste a link or write a note in the web app drag in a PDF or a doc and it gets stored, extracted, summarized image-first grid, light and dark I built this for myself, but I'm trying to work out whether it's worth turning into a proper paid app for others. That would mean a normal installer, simple local-model setup, and something I can actually support. The honest question is whether this is useful enough to buy, not whether you'd try a free version. If it were packaged properly, would you pay for it? If not, what would stop you? Local-model setup, missing mobile capture, something it doesn't do, or just not a problem you need solved? I'm not collecting emails or selling anything here. I'm trying to decide whether to productize it or leave it as my own slightly overbuilt clipping system.
已收录讨论
awesome. thanks!
Mobile capture would be a top priority in my case. It does sound promising though. I would consider paying for it as long as it’s not another subscription
I don't plan on having a monthly subscription. If that were the case, then I'd have to start worrying about cloud hosting and potentially managing the access to the AI models used for the summary in the cloud. Instead, I am thinking that this is something where all data you collect is yours and managed locally on your machine. Instead, I'm looking at a one‑time purchase that covers updates for a year, or an annual option with a reduced fee for future updates. That fee wouldn't affect the core app and only covers new features each year. If I keep adding features, you could pay the reduced amount to get next year's updates. If development stops at maintenance level, you wouldn't be charged again.
In that case, Id go for it
this looks super cool. i'd definitely purchase
same here
I'd certainly try it.
I like the idea, what model are you currently using for this?
the thing i'd test before productizing is a dirty-vault recovery run: save the same article from two devices, rename one markdown file by hand, edit its tags outside the app, then rebuild the index from scratch. it should preserve the manual edits, avoid a duplicate, and let me trace every summary back to the captured source. plain markdown is the promise i'd pay for, so the app has to treat the vault as user-owned state rather than its private database.
As an Obsidian user, I might, but it would have to be brilliant at what it does
Ollama setup wouldn't stop me buying, but it'll wreck you on support : you'll spend your life answering "why is it slow" from people on 8GB laptops. Ship a tiny default model that runs on anything, make the good one opt-in, and I'm in at a one-time $50-ish.
I’d pay for this, especially as a one-time purchase. Local Markdown and mobile capture would be the key features for me.
Yes it’s worth turning into a paid app, on both mac and mobile, the vault looks super clean well done, and I’d buy it. If it uses local AI then make it lifetime, if it uses cloud AI make it subscription based. But since people are also asking for mobile version, you might wanna use cloud AI, or local for mac/cloud for mobile
Local model setup. But you wouldn't have to with Apple's background asset feature. You could package the model that you use and upload it to Apple's servers then download it after install or during install. This is perfectly compatible with App Store sandboxing (if that is important) and you wouldn't have to give a user instructions on how to do additional setup.
I guess I would try it, but for me it is important how exactly I add new things in the vault.
I'd buy it. I think there's a bigger market than you think for Are.na or Sublime-style apps that are completely offline. I tried both but they're kind of gross - they sell you inspiration but I think really they're just mining people for data.
Let me know when it’s available.
One thing worth thinking through before productizing: the semantic search pass depends on an embedding index, and that index is inherently a derived artifact, not something a human can read or hand-edit the way the markdown files are. If the pitch is that everything lives in your vault as plain markdown and you own it, it's worth being explicit that search state sits outside that promise and gets rebuilt from the vault rather than being part of it. Otherwise you'll get bug reports that read like data loss when someone edits a file by hand and search briefly goes stale, when really the vault is fine and the index just hasn't caught up yet.
One thing worth thinking through before productizing: the semantic search pass depends on an embedding index, and that index is inherently a derived artifact, not something a human can read or hand-edit the way the markdown files are. If the pitch is that everything lives in your vault as plain markdown and you own it, it's worth being explicit that search state sits outside that promise and gets rebuilt from the vault rather than being part of it. Otherwise you'll get bug reports that read like data loss when someone edits a file by hand and search briefly goes stale, when really the vault is fine and the index just hasn't caught up yet.
One thing worth thinking through before productizing: the semantic search pass depends on an embedding index, and that index is inherently a derived artifact, not something a human can read or hand-edit the way the markdown files are. If the pitch is that everything lives in your vault as plain markdown and you own it, it's worth being explicit that search state sits outside that promise and gets rebuilt from the vault rather than being part of it. Otherwise you'll get bug reports that read like data loss when someone edits a file by hand and search briefly goes stale, when really the vault is fine and the index just hasn't caught up yet.