Prebuild Image ‐ AI HAT+ 2
This project can run completely offline on the Pi 5 with Raspberry Pi AI HAT+ 2 (Hailo-10H). You can follow the step-by-step guide to install it manually, or simply download the pre-built image and flash it onto your TF card.
- Raspberry Pi 5 (8GB or more)
- Active Cooler for Pi 5
- TF card (32GB or more)
- Raspberry Pi AI HAT+ 2 (26 TOPS, Hailo-10H)
- Whisplay HAT (screen, speaker, mic, button)
- PiSugar 3 Plus (mobile power)
Click here to download the image: whisplay-offline-aihat2-20260317.img.gz (12 GB)
Remember to extract the .gz file into an .img file before flashing it.
SSH
user: pi
password: raspberry
pisugar-power-manager
model: pisugar3
user: admin
password: admin
Note: The image may not auto-expand. It is recommended to manually expand the image to the actual size of the SD card when you log in via SSH for the first time.
sudo raspi-config --expand-rootfs
sudo rebootThis image is based on the latest version of the Raspberry Pi OS (Debian Trixie with the Raspberry Pi Desktop 64-bit, 2025-12-24) and comes pre-installed with the following software:
- hailo-h10-all (Hailo-10H system driver & firmware for AI HAT+ 2)
- hailo-apps (Hailo GenAI Python stack, required for hailo-whisper / hailo-vlm)
- Faster-Whisper (speech-to-text, running on a python server)
- hailo-ollama (LLM running on the Hailo-10H NPU, with pre-downloaded model: qwen2.5-1.5b-instruct)
- Piper (text-to-speech, running with piper-http)
- Whisplay HAT Driver (driver for the soundcard and screen)
- whisplay-ai-chatbot (this project)
- pisugar-power-manager (v2.3.2, for displaying battery level)
- sugar-wifi-conf (config Wi-Fi through BLE)
The default configuration uses Faster-Whisper (CPU) + hailo-ollama (NPU) + Piper (CPU). Since hailo-ollama exclusively locks the NPU, the ASR runs on the CPU via Faster-Whisper.
| Service | Runs on | Notes |
|---|---|---|
| Faster-Whisper (ASR) | CPU | Same as the Basic image |
| hailo-ollama (LLM) | NPU | qwen2.5-1.5b-instruct, exclusively locks NPU |
| Piper (TTS) | CPU | English model |
If you prefer to use NPU-accelerated ASR and Vision instead of NPU-accelerated LLM, you can switch to the hailo-whisper + hailo-vlm + Piper configuration. In this mode, hailo-whisper and hailo-vlm share the NPU, and the LLM falls back to CPU or a cloud API.
| Service | Runs on | Notes |
|---|---|---|
| hailo-whisper (ASR) | NPU | Can coexist with hailo-vlm |
| hailo-vlm (Vision) | NPU | Can coexist with hailo-whisper |
| Ollama / Cloud API (LLM) | CPU / Cloud | Cannot use NPU alongside hailo-whisper/vlm |
| Piper (TTS) | CPU | English model |
NPU Sharing Limitation: The Hailo-10H NPU has limited on-chip SRAM. hailo-whisper and hailo-vlm can share the NPU and run side-by-side; however, hailo-ollama exclusively locks the NPU and cannot coexist with other NPU services.
See the AI HAT+ 2 Integration Guide for detailed instructions on switching between configurations.
Use the image burning tool to flash the .img file onto a TF card (no need to configure Wi-Fi). Once the flashing is complete, insert the card into your Raspberry Pi 5 and power it on to start using it.
If you need to connect to Wi-Fi for further configuration, you can download the PiSugar APP on your mobile phone.
Use the "Wifi Config" feature to discover nearby devices, connect to your Raspberry Pi, and configure the Wi-Fi SSID and password. After a short wait, you will be able to see the Raspberry Pi's local network IP.
- If you have problems using the app, there's a web version, check out this link: https://www.pisugar.net/sugar-wifi-conf You can use the web version on android / chrome browsers.
The project folder is located at ~/whisplay-ai-chatbot. You can edit the .env file within the folder to configure the settings.
After modifying the configuration, you need to restart the service by running the following command:
sudo systemctl restart chatbot.serviceThis will apply the new settings.