On my journey to master the machine and was wondering what your setups are.
- What AI model?
- IDE or terminal?
- Pay for usage or a flat monthly amount?
Feel free to go as in-depth as you want. I’m curious to know why you chose the setup you have.
pull down to refresh
related posts
I'm primarily using Cursor's IDE with Claude 4. I think billing is flat up to a certain limit and I haven't hit the limit yet. We have a business account with Cursor.
I noticed they default to "thinking" models now. Have to go in settings to re-enable the other one.
What's your experience been like between thinking and non-thinking Claude (if any)?
Haven't tried 4 yet, but with 3.5 & 3.7 I thought non-thinking was almost always better at doing stuff, except for planning purposes. This is based entirely on vibes and it bothers me to an unnecessary extent how I can't be totally sure if my vibes are true or I'm just hallucinating/placebo'ing it somehow.
Cursor is my primary as well. Do you ever switch models or is Claude 4 good for everything?
I’ve seen headlines that say this model can do so and so for a lot cheaper and always wondered if that matters if the billing is flat.
The Good Stuff: 015 - Our Vibe Coding Setups
Media file: https://anchor.fm/s/103ad8d20/podcast/play/105875507/https%3A%2F%2Fd3ctxlq1ktw2nl.cloudfront.net%2Fstaging%2F2025-6-23%2F404391958-44100-2-7cdce25477db3.mp3
I haven't listened to this one yet but it looks highly related.
So far I haven't done much vibe coding. Would like to as I have a lot of ideas that would be nice to create. The cost and some privacy concerns are holding me back a bit.
Thanks! Found and queued it up on Fountain.
What privacy concerns do you have? Maybe give PPQ.ai a shot. You can pay with Lightning (the default is $2 worth of credits) and configure it to work with VS Code.
I just don't like the idea of big tech training their models on my data. I guess vibe coding wouldn't really be giving away too much.
I tried cypherflow where you can pay with LN or ecash with different models. I generated some code but I haven't done anything with it. I'll need to spend more time on it to find how to make it into an app or piece of software for everyone to use.
So did I
Currently testing Roo Code with
qwen3-coder. Liking Roo after I finetuned the instructions (not done yet), Qwen3 makes tons of mistakes though, so I'm not super-pleased (lots of time and compute wasted on bugfixes introducing new bugs)I had Qwen3 fix a threading / concurrent access issue it created (in golang) overnight and just found that it actually fixed it after wasting 4M+ tokens. Browsing through the iterations - it fixed the issue 14 times before wrapping a mutex solution.
Now I have to explain to Qwen that mutexes are technically antipatterns lol
This is mainly why I’m using a flat fee plan at the moment. I don’t know enough to know if a model is going in circles. 😆
Cursor with Gemini 2.5
Started tinkering with Jules on a few things (free)
I also pay for gpt for fast response, 4o, and occasional planning / research (3).
Claude free tier for copy editing.
Haven't found a terminal tool I like as a shell first user, would probably pipe that into Qwen via Groq (free tier).
Also building a thing that uses Qwen via Groq free tier.
Have you tried Claude Code for the terminal? I’ve heard good things about it.
Haven't felt Claude is worth paying for to try it, I have used the Gemini clone but it's not what I'm looking for in a terminal, I want something that runs from bash with terminal context
Typo in title
Fixed. Thanks!
good success w/ $20 cursor + BMAD
by two weeks in my progress slowed as I reached the throttle threshold for their expensive models. progress still possible, tho!
at the dayjob, I've used a fair bit of Copilot in a large production application.
I've previously used models w/ Ollama locally and Vinice.AI API to achieve some chatbots to ask development questions to.
cursor gets it done nicely.
I recommend also looking at Venice (pay by SATs for API) + OpenWebUI for personal use, enabling a local prompt interface with a hosted, nokyc inference
Did you try ppq.ai? It's also pay with lighting, per query.
I tried it 2 days ago after @BlokchainB mentioned it. It works okay-ish for me, but not with every IDE, and sometimes it hangs on a query. It's prepaid so you get dollar exposure this way, but you can top up in $2 increments.
heard of it, haven't tried it.
Honorable mention for http://shakespeare.diy/ which has been very impressive so far for quickly generating simple web apps