Vibe Coding Our AI Chat: Faster Builds, Still Human

November 1, 2025 hannesz 0

TL;DR

  • We shipped AI Chat by vibe coding—tight loops with LLMs as collaborators, not replacements.
  • It’s now radically easier to prototype complex systems; it still takes human taste, constraints, and engineering judgment.
  • Sign up and get 1,000,000 free tokens. Use the REST endpoints or the familiar chat interface—your choice.
  • Built and hosted in South Africa, powered by South African sun.

Try AI Chat Now →

Why We Built This

I found myself “vibe coding” more and more—looping with an LLM to explore a design, generate scaffolding, pressure‑test ideas, and then dropping into focused engineering to make the right calls. That rhythm made the AI Chat experience obvious: let people explore models conversationally and ship production integrations with the same stack when they’re ready.

What Changed For Me

  • From blank page to shape fast: I don’t start at zero anymore. I start with structure.
  • Less yak shaving: Docs exploration and glue‑code drafts are a prompt away.
  • Better decisions sooner: Iterating with a model surfaces constraints early.
  • Still human‑led: Naming, architecture, tradeoffs, safety, and “is this worth building?” are human jobs.

What AI Can’t Do (And That’s Fine)

  • Own the problem. AI can assist; it can’t care.
  • Carry context across weeks of messy product reality.
  • Make the hard calls when requirements collide.

I don’t think AI replaces human programmers. Developers who use AI well will outpace those who don’t. It’s a tool—powerful, yes—but still a tool.

What We Shipped

AI Chat is the conversation front‑end to the same models and GPUs behind our API. No setup. No keys. Just chat.

  • Models: Qwen3‑Coder‑30B (optimized for code + technical writing)
  • Infra: Dual NVIDIA RTX A6000s via vLLM, routed through LiteLLM
  • Same account, same credits, same behavior as your API integration

Use It Your Way

  • Explore in chat: prototype prompts, debug, learn fast.
  • Ship via API: when you’re ready, call the same model and infra.
  • Unified credits: chat and API share one balance (1,000,000 free tokens on signup).

Getting Started

A Note From the Trenches

I use LLMs daily—even for small scripts—to turn deep, multi‑system questions into concise, workable steps. The output isn’t magic; the speed is. That speed compounds.

Homegrown, Sun‑Powered

This all runs in South Africa, powered by South African sun. Local, fast, and green.

Try It