The $0 AI Software House: Running a 4-Agent Engineering Team on a $35 Raspberry Pi

This article details an experiment where a 4-agent AI engineering team (Explorer, Planner, Coder, Reviewer) runs entirely on a $35 Raspberry Pi 4 with no GPU, no API keys, and no network calls. It uses Ollama with a quantized qwen2.5-coder:1.5b model, OllaBridge for OpenAI compatibility, and GitPilot for orchestration. The author demonstrates that small models paired with a strong system can ship working applications, avoiding the high costs and dependencies of cloud APIs. The post includes a companion repo and Docker setup to simulate Pi resource limits on other machines, making it a practical guide for local, low-cost AI development.