zzzz...

think descriptive function names and doc comments are necessary for AI so it can tell what code is doing

boooring

That's totally correct, the code I've sent you had docstrings and special parameter names for a reason. That wasn't documentation but part of the functionality. With python, you can do code inspection and that include the comments. These comments goes as description for the function to the llm.

So you're goddamn right!

Power of python I guess, js can't do that I think what means that you can't build a class to json description dict to send to the llm server. With js, I would add per function name, a var with the same name with suffix _description and still use auto class to dict converter.

@retoor no in js you can do function calls. js is just using ollama REST API and it allows you to specify function calls and their descriptions -- not sure if the js lib you gave has that but the API it uses does and I use rust now and also even in js I could jerry-rig it cuz that's how js works

I don't mean function calls but if you feed coding data to the LLM and ask it to code something

... I actually didn't really understand the ragent store thing tbf. I did see you mention chromadb so I looked that up and had some snippets printed out but I think the solution like this is ugly and stupid. basically you give in the prompt relevant data from the collection. it's not really you merging the data with the model or anything. you're running the prompt input as a lookup against the DB and then pre-pending the DB search results to the AI prompt

which explains a lot about their behavior. they don't keep consistency if fed varied papers, for example

@jestdotty that's half correct! :) regarding ollama embeddings: https://ollama.com/blog/...

Is the cleanest basic tut.

@retoor which part was wrong?

this link shows the same thing

you query the collection and then in the prompt to the AI you give the search result from the collection for the AI to regurgitate at you

@jestdotty correct. I just try to help.