Hacker News new | past | comments | ask | show | jobs | submit login

I hate to do this but this is the type of app that is basically the new todo app. I create small agents like this all time in about 15 seconds, with a simple prompt like:

You are an expert Tailwind CSS developer that I will ask for help. You may need to google the latest docs from Tailwind here: link.

And I get all the help I need.




> You are an expert Tailwind CSS developer that I will ask for help.

Is that part of the prompt necessary anymore? I feel that was a version 3 thing.

My prompts have been getting less verbose & I've been finding little value in introductory phrases, correct grammar or spelling.


I think so. I notice on Meta.ai that if I ask a follow up a question, it often loses context. Cordoning it off to you are this and only this could possibly stop it from straying.


What framework/library/app do you use to execute these agents?


Just make that custom thing on OpenAi website, takes a second.

I don’t do anything locally lately, but I believe you’d either have to pad your prompt with something like what I showed you above or mess with the system prompt.

The stuff is there in the ollama docs, someone else can probably give you more details.

Better yet, just ask gpt about all this lol.

Edit: I might be misusing the word Agent here.


> Edit: I might be misusing the word Agent here.

Yeah I assumed you were using some sort of multi-agent framework where you give agents different prompts/tools and they work together. You just mean a custom GPT.

I was half hoping there was a command line app that could do that, though I also forgot custom GPTs can browse the internet.


I don’t have the hardware to run good models, but if you can run a local one then you can write a small script to set a system prompt with ollama via the command line. Haven’t tried it, but sounds like what you want.


what a great tip, thanks




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: