71–80 of 107 posts
This is a good reminder of why storing Obsidian notes as individual Markdown files is much more useful than stuffing those notes in a database and having Markdown as an export format. The direct manipulation of files allows multiple apps to coexist and do useful things on top of the same files.
i would call it bad note taking to not be able to recall an answer you put into your notes.
I like the idea of something like this but i've struggled to find a real use case.
serious question: when do you ever have your own notes and can't find the answer ? i would call it bad note taking to not be able to recall an answer you put into your notes. I like the idea of something like this but i've struggled to find a real use case.
I’m one of them.
This tool might actually make me take better notes.
Maybe.
Earlier quoted context omitted.
That was the reason why I gave up Joplin very quickly. The last Joplin thread, here on Hacker News, has also shown once again that some still do not understand why "But Joplin can export Markdown from the database!" is not the same as simple, flat Markdown files.
May I ask what you switched to? Running into the same issue.
Earlier quoted context omitted.
There are these plugins: https://github.com/zatevakhin/obsidian-local-llm https://github.com/hinterdupfinger/obsidian-ollama Which already exist and if nothing else are decent starting points. > Not sure if local LLMs are powerful enough yet to enable meaningful/reliable outcomes I've dabbled, briefly, with Ollama running Mistral locally on an M1 MacBook Pro with 32GB of unified memory, and throwing a couple of hundr…
Thanks for sharing these; I’ll definitely check these out. I somehow missed these during my initial search for similar projects. I recently got my hands on an RTX 3090 for my Linux workstation and I’m planning to try getting some kind of remote setup going for my MacBook Air. Great to hear about decent output. Reliability is negotiable as long as there’s some value and hopefully a path to future improvements.
I have been working on my note-taking application (https://github.com/dvorka/mindforger) for some time and wanted to go in the same direction. However, I gave up (for now). I used ggerganov/llama.cpp to host LLM models locally on a CPU-only machine with 32GB RAM, and used them for both RAG and note-taking use cases (like https://www.mindforger.com/index-200.html#llm). However, it did not work well for me - the performance was poor (high hardware utilization, long response times, failures, and crashes) and the actual responses were rarely useful (off-topic and impractical responses, hallucinations). I tried llama-2 7B with 4b quantization and a couple of similar models. Although I'm not happy about it, I switched to an online commercial LLM because it performs really well in terms of response quality, speed, and affordability. I frequently use the integrated LLM in my note-taking app as it can be used for many things.
Anyway, Reor "only" uses the locally hosted LLM in the generation phase of the RAG, which is a nicely constraint use case. I believe that a really lightweight LLM - I'm thinking about a tiny base model fine-tuned for summarization - could be the way to go (fast, non-hallucinating). I'm really curious to know if you have any suggestions or if you will have any in the future!
As for the vector DB, considering the resource-related problems I mentioned earlier, I was thinking about something similar to facebookresearch/faiss, which, unlike LanceDB, is not a fully-fledged vector DB. Have you made any experiments with similarity search projects or vector DBs? I would be interested in the trade-offs similar to small/large/hosted LLMs.
Overall, I think that both RAG with my personal notes as a corpus and a locally hosted generic purpose LLM for the use cases I mentioned above can take personal note-taking apps to a new level. This is the way! ;)
Good luck with your project!
Does the future of knowledge management involve using lots of AI to organize pieces of knowledge? I think "here be dragons", and that over-relying on AI to do all your organization for you will very possibly (probably?) cause you to become worse at thinking. No data to back this up because it is still early days in the proliferation of such tools, but historically making learning and thinking and "knowledge managemen…
Earlier quoted context omitted.
Files seem less useful for small bits of information. I feel the urge to fill a file with a minimum threshold. A database makes more sense for that.
>I feel the urge to fill a file with a minimum threshold. Honestly that's more you subjectively than database v files.
On the other hand, you get to implement concurrent access yourself. Multiple apps working on the same files simultaneously only works when none of them makes a mistake with caching or locking.
Super interesting project. I like its focus. Wondering if the author looked into Cozodb, or other databases that combine vector + graph/triples. Since probably neuro-symbolic is the best path. https://docs.cozodb.org/en/latest/releases/v0.6.html talks about this idea.
Earlier quoted context omitted.
That was the reason why I gave up Joplin very quickly. The last Joplin thread, here on Hacker News, has also shown once again that some still do not understand why "But Joplin can export Markdown from the database!" is not the same as simple, flat Markdown files.
May I ask what you switched to? Running into the same issue.