Back to Learn
Production
Intermediate
10 min read

The Open-Source AI Stack

Everything you need is free. Here's the map.

The Open-Source AI Stack

You can build a production AI app today without spending a cent — on free tiers and open-source software. Here's the stack we use to power LLMAtlas itself.

Models (free APIs)

  • Groq — fastest Llama 3.3 70B inference on the planet
  • OpenRouter — gateway to 30+ free model variants
  • Google AI Studio — Gemini 1.5 Flash, 1500 free requests/day
  • Cerebras — Llama 3.3 70B at record speeds
  • Cloudflare Workers AI — edge-deployed at zero cost

Local & self-hosted

  • Ollama — one-line install, runs Llama / Qwen / Phi locally
  • vLLM — production-grade serving for self-hosted
  • llama.cpp — CPU/GPU inference, even on a laptop

Embeddings & vector search

  • Nomic Embed & BGE — open embedding models
  • pgvector — Postgres extension, free with Supabase
  • ChromaDB / Qdrant — open vector DBs

App framework

  • Next.js + Tailwind + shadcn/ui — what LLMAtlas runs on
  • LangChain / LlamaIndex — orchestration
  • Promptfoo / DeepEval — evals

Hosting

  • Vercel — generous free tier
  • Cloudflare — free CDN + Workers
  • Supabase / Neon — free Postgres tiers

Total cost to launch a real AI app: $0.

Try it in the Playground

Apply what you learned with a free model.

Browse all lessons

9 lessons across 5 categories.

LLMAtlas — The Open Ecosystem Workspace for LLMs