Hermes Agent Voice AI on Raspberry Pi 4B

Complete Ubuntu Deployment Guide (STT + LLM + TTS)

This tutorial shows how to build a low-cost, always-on voice AI terminal on Raspberry Pi 4B using Hermes Agent, Whisper STT, and TTS.

---

1. Recommended Hardware

Component	Recommendation
Board	Raspberry Pi 4B (4GB or 8GB)
OS	Ubuntu Server 24.04 ARM64
Storage	SSD (strongly recommended)
Power	Official 5V 3A adapter
Cooling	Heatsink + fan
Audio	USB microphone / USB sound card

---

2. Why Not Run a Large Local LLM on Pi?

Raspberry Pi 4B has no NVIDIA GPU, so it is not ideal for:

• 7B/13B local LLM inference
• Stable Diffusion
• Large Whisper models

A practical architecture is:

• Pi handles: Hermes runtime + audio I/O + STT/TTS
• Cloud API handles: heavy LLM inference (DeepSeek/OpenAI)

---

3. Update System

sudo apt update && sudo apt upgrade -y

---

4. Install System Dependencies

sudo apt update && sudo apt install -y \
git curl wget python3 python3-pip python3-venv python3-dev \
ffmpeg portaudio19-dev build-essential cmake pkg-config \
libopenblas-dev

---

5. Create Workspace

mkdir -p ~/ai/hermes
cd ~/ai/hermes

---

6. Create Python Virtual Environment

python3 -m venv venv
source venv/bin/activate
pip install --upgrade pip

---

7. Install AI Python Packages

pip install \
faster-whisper \
edge-tts \
openai \
sounddevice \
numpy \
scipy \
requests

---

8. Install Hermes Agent

cd ~/ai
git clone https://github.com/Hermes-Project/hermes-agent.git
cd hermes-agent
pip install -r requirements.txt

---

9. Configure Whisper STT

Recommended model sizes on Pi:

• tiny (best speed)
• base (balanced)
• small (sometimes acceptable)

Quick test:

python - <<'PY'
from faster_whisper import WhisperModel
model = WhisperModel("tiny")
print("Whisper tiny loaded")
PY

---

10. Configure TTS

Install:

pip install edge-tts

Test:

edge-tts \
--text "Hello from Raspberry Pi" \
--write-media test.mp3

ffplay test.mp3

---

11. Configure LLM API (DeepSeek/OpenAI)

DeepSeek: https://platform.deepseek.com
Docs: https://api-docs.deepseek.com

Use a cloud API because Pi 4B is much better as a voice edge terminal than a heavy LLM server.

---

12. Edit Hermes Config

Typical path:

nano ~/.hermes/config.yaml

Example:

llm:
  provider: openai
  api_key: "YOUR_API_KEY"
  model: "deepseek-chat"
  base_url: "https://api.deepseek.com"

voice:
  enabled: true

stt:
  provider: faster_whisper
  model: tiny

tts:
  provider: edge_tts
  voice: en-US-GuyNeural

---

13. Start Hermes

source ~/ai/hermes/venv/bin/activate
python main.py

If your repository structure differs, try:

python app.py

---

14. Auto Start with Systemd (Optional)

Create service:

sudo nano /etc/systemd/system/hermes.service

Service file:

[Unit]
Description=Hermes Agent
After=network.target

[Service]
User=pi
WorkingDirectory=/home/pi/ai/hermes/hermes-agent
ExecStart=/home/pi/ai/hermes/hermes-agent/venv/bin/python main.py
Restart=always

[Install]
WantedBy=multi-user.target

Enable and start:

sudo systemctl daemon-reload
sudo systemctl enable hermes
sudo systemctl start hermes
systemctl status hermes

---

15. Common Issues

Build/install is very slow

Many ARM packages compile locally. This is normal.

SD card wear

AI workloads increase I/O. Prefer SSD for long-term reliability.

Out-of-memory errors

Use Whisper tiny/base, enable swap, avoid large models.

Thermal throttling

Use proper cooling and airflow.

---

16. Practical Final Architecture

Layer	Responsibility
Raspberry Pi 4B	Mic, speaker, button/control, local STT/TTS
Cloud LLM API	Reasoning and response generation
Hermes Agent	Orchestration between voice and LLM

---

17. Future Extensions

• OLED UI
• EC11 knob input
• ESP32 peripherals
• Wake-word support
• Home Assistant integration
• Telegram bot integration
• Robot arm controller endpoint

---

18. One-Command Bootstrap (Reference)

sudo apt update && sudo apt install -y \
git curl wget python3 python3-pip python3-venv python3-dev \
ffmpeg portaudio19-dev build-essential cmake pkg-config \
libopenblas-dev && \
mkdir -p ~/ai/hermes && \
cd ~/ai/hermes && \
python3 -m venv venv && \
source venv/bin/activate && \
pip install --upgrade pip && \
pip install \
faster-whisper \
edge-tts \
openai \
sounddevice \
numpy \
scipy \
requests

---

Raspberry Pi 4B is not a powerful LLM server, but it is an excellent AI voice endpoint, IoT gateway, and robotics control front-end when paired with Hermes Agent.