search

your-papa / obsidian-Smart2Brain

An Obsidian plugin to interact with your privacy focused AI-Assistant making your second brain even smarter!
GNU Affero General Public License v3.0
599 stars 38 forks source link

How to make S2B work nicely with vault #113

Open davidmarothi opened 6 months ago

davidmarothi commented 6 months ago

Feature Area

vault data

Painpoint

I really like this project, but I have troubles using it with my vault. Do you guys have any tips or tricks how to creates notes that can be used by this, or any other LLM tools? I mean, I have a query like hey, it's 2024-04-30 today. can you gather what I did yesterday? and the reply was unusable. It was pointing to nonexistent notes, and not even mentioning yesterday's daily note... how so? I'm using llama3:8B and mbai-embed-large for embedding (all running locally via ollama).

Describe your idea

Create a guide how to make LLM friendly data in our vaults. What type of data it can pick up efficiently. what kind of use cases are feasible at this time.

Alternatives

No response

Additional Context

No response

nicobrauchtgit commented 6 months ago

Hi @davidmarothi, working with time based approaches is always a bit tricky with LLMs. If you have the date in the title of your daily note it might work but it depends strongly on the model that you are using. However, we have a plan to implement a time based search algorithm to accommodate those kinds of requests without the need for a specific structure. After we're done with our exams in two weeks we'll continue to work on the plugin and put this on our agenda 😄

emory commented 3 months ago

@davidmarothi i have similar observations and it isn't clear to me that S2B is parsing the contents of my notes in my vault. it seems to know about things in PDFs filed in my vault, but the vast majority of important things in my Obsidian vault(s) are the markdown notes.

for 2-3 years i have been preparing for a system like this by really embracing dataview and an assortment of yaml attrs with the assumption it would accelerate the process of getting local model computing to be aware of the contents of my notes!