Description

ZEPIR AI Terminal - Your Pocket Sized Private AI Assistant

Private AI chatbot you can hold in your hand.

In a world dominated by cloud AI services that track your every query,

ZEPIR offers something revolutionary: your own local LLM running on this setup with a physical keyboard and display.

·        No internet required.

·        No data harvesting.

·        No monthly subscriptions.

Just you and your personal AI assistant — anywhere, anytime.

Use Cases

🤖 Offline AI Assistant (Ollama + LLM)

🌍 Multilingual Translator Translate complete sentences and paragraphs between languages. Surprisingly capable with complex phrases across many languages — not just simple words. Your personal interpreter that works offline.

💬 Casual Conversation Partner Chat naturally about various topics. Small models handle open-ended conversation well — great for language practice or just having someone to talk to.

💬 Simple Q&A Basic factual questions from training data — no internet needed. "What is the capital of Norway?"

🎤 Voice-Enabled AI (Experimental)

🗣️ Speech Recognition Voice-to-text input — speak instead of typing on the small keyboard.

🔊 Text-to-Speech Let the AI read responses aloud through the built-in speaker.

🎙️ Voice Assistant Mode Combine STT + LLM + TTS for hands-free AI interaction. Functional, but slower on smaller Pi models.

🐧 Plus: Full Linux Terminal Computer

💻 Complete Raspberry Pi OS Not just an AI gadget — it's a real Linux computer. Python, Nano, MC, Git, and thousands of packages available.

🌐 WiFi Connected Web browsing (w3m), SSH, remote server management, file downloads.

🎵 Media Playback MP3 and MOD tracker music player. Retro vibes included.

🖼️ Image Viewer Display images in framebuffer. Quick previews without desktop environment.

📡 IoT & Remote Admin SSH into servers, manage Raspberry Pis, control IoT devices from your pocket.

Features:

·        2.8" IPS 320x240 display ST7789

·        PWM sound simple amplifier mono speaker

·        optional USB-A for easy work with PI zero, 2 wires are required to test points D+,D-

·        ideal for terminal apps

·        GPIO keyboard 7x10 matrix with customizable python driver

·        comfortable FULL keyboard

No extra manufacturing, front panel is made in PCB factory + mainboard

but some 3D printed enclosure will help with that

Keeping my KISS philosophy make it stupid simple = one order = one device

ZEPIR AI Terminal — Installation Guide

Step 1: Prepare SD Card

Flash Raspberry Pi OS Lite to your SD card using Raspberry Pi Imager.

• Enable SSH

Create an empty file named ssh (no extension) in the boot partition root.

• Configure WiFi

Create file wpa_supplicant.conf in the boot partition:

Step 2: Configure Display & Audio

Edit config.txt

Add these lines to /boot/firmware/config.txt:

# Enable SPI display
dtparam=spi=on
dtoverlay=fbtft,spi0-0,st7789v,dc_pin=25,rotate=270,width=240,height=320,speed=62000000
 
# Enable audio output
dtparam=audio=on
dtoverlay=audremap,pins_18_19
audio_pwm_mode=2

Edit cmdline.txt

Add to the end of the line in /boot/firmware/cmdline.txt:

fbcon=map:10 fbcon=font:VGA8x8

Step 3: First Boot Setup

After first boot, connect via SSH and set your username and password.

Configure Console Font

sudo dpkg-reconfigure console-setup

Select: UTF-8 → Guess optimal character set → Terminus → 6x12

System Configuration

sudo raspi-config

Configure:

·        Expand filesystem

·        Enable autologin

·        Boot to terminal (CLI)

·        WiFi setup (if not done earlier)

Step 4: Test Hardware

Test Display

ls /dev/fb*
sudo dd if=/dev/urandom of=/dev/fb1

You should see random pixels on the display.

Test Audio

aplay -l
alsamixer
espeak-ng -v "Ahoj, toto je syntéza reči."

Step 5: Install Essential Packages

Update system

sudo apt update

Core utilities

sudo apt install mc w3m w3m-img mpg123 espeak-ng fbi fim python3-evdev

Extra tools & fun stuff

sudo apt install mikmod ranger nnn btop ncdu cmus mpv cmatrix sl cowsay fortune

Step 6: Setup GPIO Keyboard

Enable keyboard driver on startup

Edit /etc/rc.local and add before exit 0:

sudo python3 /home/pi/gpiokeyboard/gpiokeyboard.py &

Step 7: Install Ollama & AI Model

Install Ollama

curl -fsSL https://ollama.com/install.sh | sh

Download and Run a Model

ollama run qwen3:1.7b

This downloads the model (first time only) and starts an interactive chat session.

Recommended Models for ZEPIR

Model Size Notes qwen3:1.7b ~1.5 GB Fast, good translation capabilities ministral-3b ~2 GB Newer, solid all-round performance tinyllama ~700 MB Smallest, fastest, basic tasks gemma3:1b ~800 MB Google's lightweight model phi4-mini ~2.5 GB Microsoft's compact model

Useful Ollama Commands

# List installed models
ollama list
 
# Download model without running
ollama pull qwen3:1.7b
 
# Remove a model
ollama rm qwen3:1.7b
 
# Show model details
ollama show qwen3:1.7b
 
# Run with verbose output (see tokens/sec)
ollama run qwen3:1.7b --verbose
 
# Run with custom system prompt
ollama run qwen3:1.7b "You are a translator. Translate everything to English."
 
# Check Ollama service status
systemctl status ollama
 
# Restart Ollama service
sudo systemctl restart ollama

Tips for Small Models

·        Keep prompts short and clear

·        One task at a time works best

·        For translation: "Translate to German: [your text]"

·        Use /bye to exit chat session

·        Use /clear to reset conversation context

Voice Output

Pipe AI response directly to text-to-speech:

echo "Say hello in one sentence" | ollama run qwen3:1.7b | espeak-ng

Or with direct prompt:

ollama run qwen3:1.7b "Say hello in one sentence" --nowordwrap | espeak-ng