MIMIC AI Desktop Assistant

A privacy-first, AI-powered desktop assistant with voice synthesis, persistent memory, and local model inference. MIMIC runs entirely on your local machine using Ollama for language models and supports KittenTTS (8 selectable AI voices) and Qwen3-TTS (AI voice cloning) for voice synthesis.

Latest Release: v1.2.0 - See Release Notes for new features!

Table of Contents

• Features
• Video Showcase
• Release Notes
• System Requirements
• Installation
• Quick Start
• Architecture Overview
• Configuration
• Usage Guide
• Troubleshooting
• Privacy & Security
• License

---

Video Showcase

Click to watch MIMIC AI in action - Voice cloning, multi-persona support, and real-time conversation

🎥 Watch the Demo: MIMIC AI Desktop Assistant v1.2.0

---

Features

Core Capabilities

• Local AI Inference: Uses Ollama for running LLMs locally (Llama, Mistral, etc.)
• Voice Synthesis:
  • KittenTTS (Default): 8 selectable AI voices (Bella, Jasper, Luna, Bruno, Rosie, Hugo, Kiki, Leo) with adjustable speed
  • Qwen3-TTS: AI voice cloning with reference audio (~3-6GB VRAM)
• Persistent Memory: Conversation memory with per-persona context and file-based storage
• Web Search: Privacy-focused SearXNG integration with smart result summarization
• Wake Word Detection: Hands-free activation with custom wake words
• Vision Support: Image analysis with vision-capable models
• Audio Watermarking: All generated audio is invisibly watermarked for identification

Persona System

• Create multiple AI personas with unique personalities
• Per-Persona Memory Isolation: Each persona has their own memory folder (~/MimicAI/Memories/{persona_id}/)
• Full conversation history stored per persona
• Voice creation with adjustable parameters (pitch, speed, emotion, etc.)
• Avatar customization with dynamic 3D visuals

Avatar Personality System

• Automatic Trait Extraction: Derives avatar personality traits from personality prompts
• Voice Parameter Derivation: Automatically adjusts voice parameters based on personality
• Dynamic Emotional State: Avatar tracks emotional states during conversations
• Height-Based Camera: Camera intelligently positions based on avatar model height
• Procedural Vocalizations: AI-generated non-verbal sounds (giggles, sighs, hums)

Smart Router System

• Intent Classification: Lightweight LLM routes queries to appropriate handlers
• Search Summarization: Router automatically summarizes web search results to reduce token usage
• Minimal System Prompts: System prompts stay under 500 tokens for faster inference
• No Emoji Policy: Responses exclude emojis (TTS reads them aloud)

Memory Management (Improved in v1.2.0)

• Per-Persona Storage: Each persona gets isolated memory folder with:
  • user_files/ - Uploaded documents
  • conversations/ - Saved conversation files
  • history.json - Full conversation history
• AI Memories: Important conversation points automatically extracted
• Full History: Complete conversation log with search capability
• No File Size Limits: Upload any size file (user-managed local storage)

Licensing

• Freemium Model: Core functionality free, premium assets available in-store
• Proprietary License: See LICENSE.txt for full EULA
• Contact: deadheadstudios@atomicmail.io

---

Release Notes

v1.2.0 (Latest)

Major Changes

• 🔊 KittenTTS Replaces Browser TTS: Default TTS is now KittenTTS with 8 selectable voices
  • Female: Bella, Luna, Rosie, Kiki
  • Male: Jasper, Bruno, Hugo, Leo
  • Adjustable speech speed (0.5x - 2.0x)
• 🧠 Improved Memory System: Per-persona memory folders with full conversation history
• 🎯 Smart Router: Intent routing and automatic search result summarization
• 📷 Camera Improvements: Better avatar following and height-based positioning
• 🧹 Clean Console: Removed debug logging for production builds
• 📝 Minimal System Prompts: Reduced token usage for faster responses
• 🚫 No Emojis: Responses exclude emojis for better TTS experience

Technical Improvements

• File size limits removed (user-managed storage)
• Search results processed by router to reduce context size
• AI Memories tab shows current persona only
• Voice selection persists across sessions

---

System Requirements

Minimum Requirements

• OS: Windows 10/11, macOS 12+, or Linux
• RAM: 8GB (16GB recommended)
• Storage: 5GB free space
• GPU: Optional but recommended for faster inference

Software Dependencies

• Python: 3.10, 3.11, or 3.12 (3.13+ not supported due to PyTorch compatibility)
• Node.js: 18.x or 20.x LTS
• Ollama: Latest version from ollama.com
• Docker Desktop: Optional, for SearXNG web search

VRAM Requirements (Voice Synthesis)

• KittenTTS: No VRAM required (CPU-based, 15M-80M parameter models)
• Qwen3-TTS 0.6B: ~3GB VRAM
• Qwen3-TTS 1.7B: ~6GB VRAM

---

Installation

Desktop Application (Recommended)

Run the installer:

.\Mimic AI_1.0.0_x64-setup.exe

What happens:

• Installs MIMIC to your system
• Creates Start Menu and Desktop shortcuts
• All services run in background (no terminal windows)
• Auto-starts/stops Ollama, TTS backend, and SearXNG

---

Quick Start

First Launch

1. Accept the End User License Agreement (EULA)
2. The app will auto-detect Ollama models
3. Create your first persona or use the default "Mimic" assistant

Creating a Voice (KittenTTS - Default)

1. Go to Voice Studio tab
2. Select or create a persona
3. Choose KittenTTS engine
4. Select one of 8 voices (Bella, Jasper, Luna, etc.)
5. Adjust speech speed if desired
6. Save to persona

Creating a Voice (Qwen3-TTS - AI Cloning)

1. Go to Voice Studio tab
2. Select or create a persona
3. Choose Qwen3-TTS engine
4. Upload reference audio or record your voice
5. Adjust voice parameters (pitch, speed, emotion)
6. Click "Create Voice" and save to persona

Using Web Search

1. Enable "Web Search" in Settings
2. Docker Desktop will auto-start if installed
3. SearXNG container will be created and started
4. Ask questions like "What's the weather today?"
5. Router automatically summarizes results

---

Architecture Overview

System Architecture

┌─────────────────────────────────────────────────────────────┐
│                        Frontend (React)                      │
│  ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐       │
│  │  Chat    │ │ Settings │ │  Voice   │ │ Personas │       │
│  │  Panel   │ │  Panel   │ │  Studio  │ │ Manager  │       │
│  └──────────┘ └──────────┘ └──────────┘ └──────────┘       │
│                         │                                    │
│              ┌──────────┴──────────┐                        │
│              │     Zustand Store    │                        │
│              │  (State Management)  │                        │
│              └──────────┬──────────┘                        │
└─────────────────────────┼───────────────────────────────────┘
                          │
┌─────────────────────────┼───────────────────────────────────┐
│              TTS Backend (Python FastAPI)                    │
│  ┌─────────────┐ ┌─────────────┐ ┌─────────────────────┐   │
│  │  KittenTTS  │ │  Qwen3-TTS  │ │  Legal Watermarker  │   │
│  │   Engine    │ │   Engine    │ │    (AudioSeal)      │   │
│  └─────────────┘ └─────────────┘ └─────────────────────┘   │
│                         │                                    │
│              ┌──────────┴──────────┐                        │
│              │   Unified TTS Server  │                      │
│              │     (Port 8000)       │                      │
│              └───────────────────────┘                      │
└─────────────────────────────────────────────────────────────┘
                          │
┌─────────────────────────┼───────────────────────────────────┐
│              Ollama Server (Local LLM)                       │
│  ┌─────────────┐ ┌─────────────┐ ┌─────────────────────┐   │
│  │   Llama     │ │   Mistral   │ │   Vision Models     │   │
│  │   Models    │ │   Models    │ │  (LLaVA, BakLLaVA)  │   │
│  └─────────────┘ └─────────────┘ └─────────────────────┘   │
│                         │                                    │
│              Port 11434 (Default)                           │
└─────────────────────────────────────────────────────────────┘

Data Storage (Per-Persona in v1.2.0)

┌─────────────────────────────────────────────────────────────┐
│                    Data Persistence                          │
├─────────────────────────────────────────────────────────────┤
│  localStorage (Browser)                                     │
│  ├── mimic_personas       - Persona configurations          │
│  ├── mimic_settings       - App settings                    │
│  └── mimic_app_state      - Current state                   │
│                                                             │
│  IndexedDB (Browser)                                        │
│  └── Voice audio data (large binary data)                   │
│                                                             │
│  File System (~/MimicAI/Memories/{persona_id}/)             │
│  ├── user_files/          - Uploaded documents              │
│  ├── conversations/       - Conversation exports            │
│  └── history.json         - Full conversation history       │
└─────────────────────────────────────────────────────────────┘

---

Configuration

Settings

Settings are stored in localStorage under mimic_settings:

{
  "ollama_url": "http://localhost:11434",
  "default_model": "llama3.2",
  "vision_model": "bakllava",
  "tts_engine": "kitten",
  "kitten_voice": "Bella",
  "kitten_model": "nano",
  "kitten_speed": 1.0,
  "voice_volume": 1.0,
  "enable_memory": true,
  "enable_web_search": false
}

---

Usage Guide

Chat Interface

1. Text Input: Type messages and press Enter
2. Voice Input: Click microphone icon or use wake word
3. File Attachment: Click paperclip to upload files (no size limit)
4. Image Upload: Click camera icon for vision analysis

Persona Management

1. Go to Personas tab
2. Click "New Persona" to create
3. Configure name, personality, wake words, avatar
4. Each persona gets isolated memory storage

Memory Manager

1. Go to Memory Manager
2. Memory Files: Manage uploaded documents (per-persona)
3. AI Memories: View important extracted memories
4. Full History: Complete conversation log

Voice Studio

1. Select persona to voice
2. Choose engine:
   • KittenTTS: Select from 8 voices, adjust speed
   • Qwen3-TTS: Upload reference audio for cloning
3. Adjust parameters (pitch, speed, emotion)
4. Save to persona

---

Troubleshooting

Common Issues

"Python not found" error

• Install Python 3.10-3.12 from python.org
• Ensure "Add to PATH" is checked

"Ollama not connected"

• Verify Ollama is running: ollama serve
• Check Ollama URL in Settings
• Ensure at least one model is pulled: ollama pull llama3.2

"TTS backend not connected"

• Check if backend is running on port 8000
• Restart the application
• Check firewall settings

No audio output

• Check system volume
• Verify voice is saved to persona
• Try different TTS engine (KittenTTS requires no setup)

Web search not working

• Ensure Docker Desktop is installed
• Check if SearXNG container is running: docker ps
• Verify port 8080 is not in use

---

Privacy & Security

Data Locality

• All AI processing runs locally - No cloud services required
• Voice data never leaves your machine
• Chat history stored locally in per-persona folders

Audio Watermarking

All AI-generated audio contains invisible watermarks for:

• Transparency: Identifies AI-generated content
• Legal protection: Forensic evidence of origin
• Misuse prevention: Deters fraudulent use

Per-Persona Memory Isolation

Each persona's data is stored in separate folders:

• No cross-contamination between personas
• Easy backup/restore per persona
• Delete one persona without affecting others

---

Media Gallery

Adding Demo Media to README

You can embed demo media in this README using GitHub's file hosting:

Option 1: GitHub Repository Media (Recommended)

1. Create a media/ folder in your repository
2. Add GIFs/videos to the folder
3. Reference them in README:

   ![Demo](media/demo.gif)
   

Option 2: GitHub Issue Attachments

1. Create a new GitHub issue
2. Drag and drop your GIF/video into the issue
3. Copy the generated URL
4. Use that URL in your README

Option 3: GitHub Repository Assets

1. Go to your repository's "Releases" section
2. Create a new release
3. Upload GIFs/videos as release assets
4. Link to them in README

Recommended Format:

• GIFs: Use for short demos (under 10 seconds) - autoplay in README
• Videos: Use MP4 for longer demos - users click to play

---

Attribution

Character Model

• Default Avatar: PeePhanthong (VRoid Hub)

Animations

• VRoid Project by pixiv Inc.
• 八ツ橋まろんのお店 (Yatsuhashi Maron's Shop)
• Kannaku @ Nekokoya

See ATTRIBUTION.md for full credits and license terms.

---

License

Proprietary License - See LICENSE.txt

Copyright (c) 2026 Dead Head Studios. All Rights Reserved.

This software is provided under a Proprietary License. The Software is licensed, not sold. See the EULA for complete terms including:

• Freemium model with optional premium content
• Usage restrictions
• Intellectual property rights
• Termination conditions

Contact: deadheadstudios@atomicmail.io

---

Support

• Email: deadheadstudios@atomicmail.io
• Web: https://github.com/bmerriott/MIMIC-Multipurpose-Intelligent-Molecular-Information-Catalyst-
• Documentation: See QUICKSTART.md