Show HN: AI agents running on 2011 Raspberry Pi with pure PHP – no GPU (github.com)

Datapizza-AI PHP is an ultra‑minimal, educational port of Datapizza AI rewritten in pure PHP 7.x to run on tiny hardware — think a 2011 Raspberry Pi Model B (≈3W) or any shared LAMP/XAMPP host. It deliberately avoids GPUs, containers, package managers and heavy SDKs: everything is hand‑coded, file‑based and API‑first so developers can see every embedding, cosine calculation and retrieval step. The project exposes a complete RAG + agent pipeline (embedders, JSON vector store, retrieval, agents with tools and simple memory) and ships with examples, a tiny HTTP server, and integrations (n8n, Model Context Protocol). Technically, it uses remote text embeddings (text-embedding-3-small), stores vectors in plain JSON (/data/vectors.json), computes cosine similarity in vanilla PHP ($dot / (sqrt($na)*sqrt($nb))), and performs single‑threaded file I/O with simple locks and an SHA‑1 embedding cache. Requirements are modest (PHP ≥7.0, curl and json, ~256MB RAM); no DB, no Composer, optional SQLite. It’s not enterprise-grade — remote API calls are required, and performance won’t match GPU-backed stacks — but it’s significant for the AI/ML community as a teachable, auditable reference implementation: perfect for local document search, home automation assistants, privacy‑first prototyping, and teaching retrieval/agent internals without hiding the math.