
Quick one today.
I've been building side projects for a while now, and every time I wanted to add "chat with AI" to something, my brain went straight to OpenAI's API — which means straight to a bill.
Turns out that's not necessary anymore, at least not for prototyping or small projects.
Groq gives you free access to solid open models (GPT-OSS, Llama, etc.) — and here's the part that actually surprised me: their API is fully OpenAI-compatible. Not "similar," actually compatible. You install the regular openai npm package, point baseURL at Groq's servers instead of OpenAI's, and everything else — chat.completions.create, the response shape, all of it — just works.
No new SDK to learn. No separate integration pattern. Just a different address.
I recorded the whole thing start to finish — creating a Next.js app, hitting Groq with a raw fetch first (so you actually see what's happening instead of trusting a black-box SDK call), then swapping to the OpenAI SDK, then wiring up a basic UI on top of it.
Video's here if it's useful: How To Use Free AI Models In Your Application (Groq + Next.js)
Couple things worth knowing if you go this route:
— Free tier means rate limits. I hit one on camera in the video — not something you want finding out in production.
— This isn't just a Groq thing either. Mistral, OpenRouter, Hugging Face — most of the free-tier providers out there have gone OpenAI-compatible too, so this pattern isn't a one-off trick, it's becoming the norm.
Next thing I'm building (and will probably post about here) — automatically failing over between providers when one rate-limits you, so a demo or small app doesn't just die mid-request. Using a tool called Omniroute for that.
Curious if anyone here's already doing multi-provider failover in production, or if you're all just eating the OpenAI bill and moving on. Would love to know what's actually working for people at the "I have 50 users and can't justify a $200/mo API bill yet" stage.