Your Pi Zero becomes a pocket AI walkie-talkie

What it does

This is the software stack for Whisplay, a pocket-sized AI gadget built around the Raspberry Pi Zero 2W or Pi 5. Press a button, speak into the onboard microphone, and the device speaks a response through its built-in speaker while showing output on a small LCD. The project pairs with PiSugar’s Whisplay HAT, battery board, and 3D-printable enclosures to create a self-contained, walkie-talkie-style chatbot.

The interesting bit

Most voice assistants are thin clients streaming to the cloud; this one is designed to run its entire ASR, LLM, and TTS pipeline locally when equipped with a Pi 5, 8 GB of RAM, and an AI accelerator like the LLM8850 or Raspberry Pi AI HAT+ 2. That turns a gadget you can carry in a pocket into a fully private, offline conversational device.

Key highlights

• Offline voice pipeline: supports local ASR, TTS, and LLM inference via the LLM8850 or Hailo-10H accelerators, plus multimodal Qwen3-VL and image generation on supported hardware
• Push-to-talk and optional wake-word interaction through an integrated speaker, microphone, and LCD bundled on the Whisplay HAT
• Pre-built OS images and ready-to-print enclosure files for both Pi Zero 2W and Pi 5 form factors
• TypeScript and Node-based core with Python backend components, all under GPL-3.0

Caveats

• The README is heavy on quickstart steps and light on architecture or performance benchmarks.
• Offline operation is explicitly recommended only for the Pi 5 with 8 GB RAM and specific accelerator cards; local inference on the Pi Zero 2W is not documented.
• Tool use is explicitly unsupported for the LLM8850 Qwen3 and Qwen3-VL integrations.

Verdict

A solid fit for hardware hackers already in the PiSugar ecosystem who want a tangible, offline-capable voice AI device. Skip it if you are looking for a generic software-only assistant or a polished consumer product.