Back to Downloads

Local Unity AI

v2.5

A completely local AI assistant that runs entirely on your computer. Chat with Mistral, Llama 3, Qwen 2.5, or Phi-3. Generate images with SDXL models. No internet required. No API keys. 100% private. Your data never leaves your machine.

Download Local Unity v2.5.zip
File Size ~50 MB
Platform Windows/Linux
License Free
Last Updated December 2025

Screenshots

AI Model Download Wizard Click to expand

Model Download Wizard - Select text models (Llama, Mistral, Qwen, Phi) and image models (DreamShaper, epiCRealism, Juggernaut)

Local Unity AI Chat Interface Click to expand

Full chat interface with AI-generated image, console output, sessions panel, and settings

Table of Contents

Overview & Features

Local Unity AI brings the power of large language models and image generation directly to your computer. Everything runs on YOUR hardware using YOUR graphics card. No subscriptions, no API keys, no cloud services, no data collection. Complete privacy.

  • Local LLM Chat
    Chat with Mistral 7B, Llama 3.2, Qwen 2.5, or Phi-3. All models run locally via llama-cpp-python with CUDA acceleration.
  • SDXL Image Generation
    Generate images with ponyRealism, DreamShaper, epiCRealism, Juggernaut XL, and more. Photorealistic or artistic styles.
  • Model Download Wizard
    First-run wizard lets you select which models to download. Only install what you need. Add more later from Hugging Face or CivitAI.
  • Session Management
    Save and organize chat history. Pick up conversations where you left off. Multiple sessions supported.
  • Memory System
    AI remembers context across sessions. Persistent memories stored locally in JSON files.
  • Customizable Personality
    Edit the system prompt in data/system_prompt.md to change the AI's personality and behavior.
  • Voice Output (TTS)
    AI can speak responses aloud. Dark gothic hacker-themed interface.
  • One-Click Startup
    Double-click run.bat (Windows) or ./start.sh (Linux). Browser opens automatically.

System Requirements

Operating System

Windows 10/11 (64-bit) or Linux with CUDA

Python

Python 3.10 or newer (check "Add to PATH")

Graphics Card

NVIDIA GPU with CUDA (GTX 1060 6GB+, RTX recommended)

System RAM

16 GB minimum, 32 GB recommended

Storage

20-50 GB for models (depends on selection)

CUDA Toolkit

CUDA 12.x recommended (auto-detected)

GPU Required

Local Unity AI requires an NVIDIA GPU with CUDA support. AMD GPUs are not currently supported. CPU-only mode is extremely slow and not recommended.

Installation

1

Install Prerequisites

2

Download & Extract

Download the Local Unity v2.5.zip file and extract it to a simple path like C:\AI\LocalUnity. Avoid paths with spaces or special characters.

3

Run First-Time Setup

Windows: Double-click run.bat

Linux: Run chmod +x start.sh && ./start.sh --setup

First run will automatically:

  • Create a Python virtual environment
  • Install PyTorch with CUDA support
  • Install llama-cpp-python for text models
  • Install diffusers for image generation
  • Show the Model Download Wizard

This takes 5-10 minutes the first time. Subsequent starts are instant.

4

Select Models

The Model Download Wizard will open in your browser. Select at least one text model and optionally an image model. Recommended for most users:

  • Text: Mistral 7B Instruct (4.1 GB) - Best balance of quality and speed
  • Image: ponyRealism V2.2 (7.1 GB) - Great photorealistic results
5

Start Chatting!

Once models are downloaded, the chat interface opens at http://localhost:8080. Type a message and press Enter. For image generation, just describe what you want and enable the image toggle.

Available Models

Text Models (GGUF Format)

Model Size Description
Mistral 7B Instruct v0.3 4.1 GB Recommended. Excellent all-around chat model.
Llama 3.2 3B Instruct 2 GB Meta's compact model. Fast and efficient.
Llama 3.1 8B Instruct 4.9 GB Meta's flagship. High quality responses.
Phi-3.5 Mini Instruct 2.4 GB Microsoft's efficient small model.
Qwen 2.5 7B Instruct 4.7 GB Alibaba's model. Great for coding and multilingual.

Image Models (SDXL Safetensors)

Model Size Style
ponyRealism V2.2 7.1 GB Photorealistic. Great faces and details.
DreamShaper XL v2 Turbo 6.9 GB Fast artistic/creative generations.
epiCRealism XL v5 6.9 GB Ultra realistic photography style.
Juggernaut XL v9 6.9 GB Most downloaded SDXL model. Versatile.

Using the Interface

Chat

Type your message in the input field at the bottom and press Enter or click Send. The AI will respond using your selected text model. Responses stream in real-time.

Image Generation

Enable the IMAGE toggle in settings, then describe what you want to generate in chat. The AI will create an image based on your description. Images appear inline in the chat.

Sessions

Click SESSIONS to view, create, or switch between conversation sessions. Each session maintains its own chat history. Sessions are saved to data/sessions.json.

Settings

Access SETTINGS to change models, enable/disable features, adjust TTS voice, and more. You can also edit data/system_prompt.md to customize the AI's personality.

Adding Custom Models

Text Models (GGUF)

Download .gguf files from Hugging Face and place them in the models/ folder:

models/
├── mistral/
│ └── Mistral-7B-Instruct-v0.3.Q4_K_M.gguf
├── llama3/
│ └── Llama-3.2-3B-Instruct-Q4_K_M.gguf
└── your-model/
└── YourModel.gguf

Image Models (Safetensors)

Download .safetensors files from CivitAI or Hugging Face and place them in models/image/:

models/image/
├── ponyRealism_V22/
│ └── ponyRealism_v22MainVAE.safetensors
├── your-model/
│ └── YourModel.safetensors

Model Recommendations

Look for Q4_K_M or Q5_K_M quantized GGUF models for best balance of quality and speed. For images, SDXL-based models work best. Check CivitAI ratings and samples before downloading.

Ready for Local AI?

Download Local Unity AI and start chatting with AI models running entirely on your hardware. Complete privacy. No subscriptions. No limits.

Download Local Unity v2.5.zip

Need help? Join our Discord community