**You Can Turn Your Raspberry Pi Into A Local AI Agent - Here's How**

Turn your Raspberry Pi into a local AI agent with this step-by-step guide

Raspberry Pi enthusiasts, rejoice! The possibilities of what you can do with one of these tiny powerhouses seem endless. From transforming your smart home to upgrading your bedroom, we've covered it all before. But one project takes the cake - Simone Marzulli's local AI agent built on a Raspberry Pi 5. With no external services in sight, this project is a masterclass in safeguarding user data and pushing the boundaries of what's possible.

Marzulli's goal was simple: keep everything contained within the Raspberry Pi. No offloading processes to external services, no compromises on security. And that meant powering his AI bot with open large language models (LLMs). He also wanted voice control - a wake word would trigger the system, and users could interact with it using a touchscreen or voice commands.

With these requirements in mind, Marzulli built Max Headbox, a screen-based AI bot that's equal parts assistant and display. Using GIMP to animate Microsoft's Fluent Emojis, he created a face that reacts to voice commands. The system is also equipped with touchscreen functionalities, where users can tap to cue the microphone or cancel requests.

For his LLMs, Marzulli chose Qwen3 1.7b for the agentic model (decision-making) and Gemma3 1b for the conversational model (emotional reactions). Given the hardware limitations of the Raspberry Pi, he had to be strategic in choosing models that wouldn't tank the CPU and memory. He opted for models within the 1B–2B parameter range for a balance between speed and expressive capability.

**How to Build Your Own Max Headbox**

If you're itching to build your own AI companion, Simone Marzulli has documented his project on GitHub with parts lists, setup instructions, and configuration details. Here's what you'll need:

  • Raspberry Pi 5 (tested on 16 GB and 8 GB models)
  • USB microphone (recommended)
  • GeeekPi Screen, Case, and Cooler
  • Ruby 3.3.0, Node 22, Python 3, and Ollama installed on the Pi
  • Vosk API for wake-word detection and faster-whisper for speech transcription

For voice control, Marzulli relied on the Vosk API to handle wake-word detection and faster-whisper for fast, accurate speech transcription. When it comes to building tools the agent can use, the process is straightforward:

  • Create a JavaScript module that exports an object with four fields: tool name, parameters accepted, brief description, and function executing the task

If you've got some free time on your hands and enjoy a good DIY challenge, building your own Pi-powered AI companion is absolutely within reach. And for more Raspberry Pi inspiration, check out our roundup of the best Raspberry Pi projects to try in 2025.

**Conclusion**

Simone Marzulli's local AI agent built on a Raspberry Pi 5 is an impressive achievement that showcases the potential of this tiny powerhouse. With its open source nature, users can modify and expand upon the project to create their own unique AI companions. Whether you're a seasoned DIY enthusiast or just starting out with Raspberry Pi projects, Max Headbox is an inspiring example of what's possible when creativity meets technology.