Built Datapizza-AI in PHP on 2011 Raspberry Pi (256MB RAM) — Zero GPU
I ported Datapizza-AI from Python to PHP to explore edge AI on ultra-constrained hardware. This runs on a 2011 Raspberry Pi Model B with only 256MB of RAM.
The Challenge
- Ultra-constrained hardware (256MB RAM, 700MHz CPU)
- Zero heavy dependencies — just curl + OpenAI API
- No local ML models, no GPU
The Result
Fully functional AI agents using ReAct pattern with:
- Tool-calling and reasoning
- Multi-model support (OpenAI, DeepSeek, Mistral, Anthropic, Kimi)
- Vector search with JSON-based storage
- Full streaming support
Why This Matters
Proves that intelligent API architecture beats raw compute. You don't need expensive hardware when you design systems API-first.
The Project
📦 GitHub: https://github.com/paolomulas/datapizza-ai-php
📝 Full Technical Breakdown: https://medium.com/@paolomulas_29555/the-smallest-ai-lab-in-the-world-built-with-php-and-a-raspberry-pi-6decc45639e6
Key Features
- ReAct agents with tool-calling
- Vector search with JSON persistence
- Streaming responses
- Modern PHP 7.x+ code
- Framework-agnostic
- ~150 lines to get started
Curious if anyone else is doing edge AI experiments on legacy hardware! Happy to discuss implementation details.
Top comments (0)