FSRS just works, even without a GPU so it's not the cool kind of AI / machine learning these days.
No joke though: the FSRS model is marvelous, and Anki remains one of the best free + open source implementations around.
I've been learning German recently and Anki (in FSRS mode) is one of the most important learning tools I have. No joke.
------
Every card remembers every rating you give it, as well as the time / date. This allows for Anki to solve for a 'forgetting curve', and predict when different cards have a chance to be forgotten.
There is furthermore the machine learning / stochastic descent algorithm to better fit the assumed forgetting curves to your historical performance. This is the FSRS Optimize parameters button in the settings panel.
> Every card remembers every rating you give it, as well as the time / date. This allows for Anki to solve for a 'forgetting curve', and predict when different cards have a chance to be forgotten.
True to a point; every card has its ratings, but the "forgetting curve" algo of FSRS is only tuned to the deck (or "option set") that the card is in, not per card.
The entire FSRS parameter set (~20+ parameters, depending on FSRS version) is per deck.
Each card is tuned to... 2 parameters IIRC? f(Difficulty, Stability, Time) == Retrievability. Time is just time so its not really a parameter, but Difficulty and Stability is solved on a per-card basis.
If there was an easy way for productivity apps to do that, it would also be a good way for malware to do that. It could also still be tricked, for example, by changing the system date on your device.
I also use Karakeep, but for me the value is in the full text search + AI tagging. I very often remember some page or article I visited a while ago and want to revisit it, only to not be able to find it because I don't remember where it was. Karakeep essentially acts as my own personal search engine to let me search through the best pages I visited.
That makes sense — treating it as a personal search engine is a real, high-ROI use case. Full-text search covers the “I remember the idea but not where I saw it” problem really well.
Out of curiosity, what’s the bigger win for you: full-text search itself, or the tagging/metadata layer that helps narrow results when your memory is fuzzy? And do you mostly search by keywords, or by “context” (project/topic you’re working on)?
I’m validating a similar retrieval-first angle (summarized in my HN profile/bio if you want to compare notes).
Given that your comment is AI generated I don't know if you're actually interested or just want to plug your product, though I'll assume good faith and answer the question
I don't manually tag any entries - the automatic AI tags just add extra keywords I can search for that are not included in the original article text. So I mostly search by keywords, yes. Not sure what the difference is between "keywords" and "topic you're working on".
See also https://mymind.com, which takes the AI tagging even further. Potentially similar to what you're building (although, again, your landing page contains a lot of AI generated metaphors and nothing that explains what your product actually does)
As mentioned in my current Ask HN post, the product is indeed not yet finalized or launched. The envisioned product, Concerns, acts as a bridge, linking your current concerns and target tasks to your knowledge base/resource repository and action list (which could be to-do lists, calendars, etc.), forming an organic closed-loop system. Using target/active projects within Concerns as triggers, it retrieves relevant information from your resource repository. It proactively pushes solutions, plans, and suggestions for users to filter. Selected items then enter the user's action list. The goal is to enhance efficiency and effectiveness in a lightweight manner, without altering existing habits for using resource repositories or action lists.
This idea stems from my own pain points, and I genuinely hope that while solving my own issues, it might also address broader needs.
Regarding your response: It's interesting that AI tagging primarily aids by adding extra searchable keywords. However, I'd prefer broader content and semantic search/matching capabilities without relying solely on tags—though tagging remains a viable implementation approach.
Thanks for the mymind reference—I'll explore it.
PS. Did you perceive an AI-driven approach because I used translation software?
> PS. Did you perceive an AI-driven approach because I used translation software?
Are you using an LLM-based translation tool? I perceived your comment as AI mostly based on the first paragraph:
> That makes sense — treating it as a personal search engine is a real, high-ROI use case. Full-text search covers the “I remember the idea but not where I saw it” problem really well.
This is very much an LLM-style "That's a great idea!" type response. I usually don't even notice when something is LLM generated, but this part really stood out even to me.
It seems like most software now integrates LLM... Can't escape it, sigh.
The mymind you recommended has made significant strides toward tackling “information overload” and “organization fatigue.” However, I feel it remains fundamentally a storage solution—reducing the effort of organizing and facilitating retrieval—but doesn't directly align with my target.
It also reminds me of another product, youmind (https://youmind.com/), though it's primarily geared toward creation rather than PKM. Perhaps I could pay to try its advanced AI features.
> However, task management apps are so unbelievably common nowadays.
And yet, there are still basically no good task management apps for desktop. Todoist is the only one that comes to mind, but it's closed source and cloud/subscription based.
I'm really waiting for the desktop tasks.org client to come out. Until then, I can only manage my tasks from my phone, because no other FOSS apps come even close.
There is a jewelry store in my country which constantly used this tactic and had a "closing soon" sale for YEARS, to the point where it became a meme associated with that specific company. Then they launched a new marketing campaign with the slogan "this time we're not closing - but our prices are as low as if we were!"
Maybe they messed something up in the official interface then. I've heard that the PDF processing capabilities are also significantly worse in Gemini UI compared to using it through the API or Google AI Studio.
