.S sSSSSs .S_sSSs .S sSSs .S_SSSs .S
.SS d%%%%SP .SS~YS%%b .SS d%%SP .SS~SSSSS .SS
S%S d%S' S%S `S%b S%S d%S' S%S SSSS S%S
S%S S%S S%S S%S S%S S%| S%S S%S S%S
S&S S&S S%S S&S S&S S&S S%S SSSS%S S&S
S&S S&S S&S S&S S&S Y&Ss S&S SSS%S S&S
S&S S&S S&S S&S S&S `S&&S S&S S&S S&S
S&S S&S sSSs S&S S&S S&S `S*S S&S S&S S&S
S*S S*b `S%% S*S S*S S*S l*S S*S S&S S*S
S*S S*S S% S*S S*S S*S .S*P S*S S*S S*S
S*S SS_sSSS S*S S*S S*S sSS*S S*S S*S S*S
S*S Y~YSSY S*S SSS S*S YSS' SSS S*S S*S
SP SP SP SP SP
Y Y Y Y Y
A locally-running AI chat assistant with memory systems, personality features, and modular architecture. Built to run entirely on your hardware with complete privacy.
I do not support non-local AI services. I believe they are an insult to individual privacy, autonomy, and intelligence.
Cloud-based AI services treat users as products, harvesting personal data, conversations, and behavioral patterns for corporate profit. They create dependency on external infrastructure, compromise privacy, and undermine human agency in the AI relationship.
My goal with Ignis is to show that good AI can be built using open-source efforts - running entirely on local hardware, under your complete control.
Ignis represents an approach to AI that's private and personal. No data collection, no surveillance, no corporate middlemen. Just you and an AI that respects your privacy.
Join the local AI revolution. Take back control of your conversations and your data.
Ignis AI combines several useful features for local AI:
- 🧠 Memory Systems: Multi-layered memory with different types for conversations and knowledge
- 🎭 Personality Features: AI that adapts communication style through interactions
- ⚡ Good Performance: Efficient design with GPU support when available
- 🔒 Complete Privacy: 100% local processing - no data leaves your machine
- 🎨 Nice Interface: Clean chat interface with real-time responses
- 🔧 Extensible Design: Plugin system for adding functionality
Experience AI that thinks, remembers, and evolves - all running on your local machine.
Double-click Ignis.bat to launch the web interface. The chat will automatically open in your default browser!
That's it! Your advanced AI assistant is ready to chat with full memory and personality capabilities.
Ignis includes a multi-layered memory system for conversations and knowledge:
- Episodic Memory: Remembers specific events and conversations
- Semantic Memory: Stores facts and relationships
- Working Memory: Manages current conversation context
- Procedural Memory: Learns interaction patterns
- Ebbinghaus Curves: Scientific memory decay modeling
- Spaced Repetition: Review scheduling for better retention
- Interference Modeling: Prevents memory confusion
- Memory Consolidation: Strengthens important memories
- Semantic Search: Finds relevant information using meaning
- Confidence Scoring: Rates memory relevance
- Source Tracking: Knows where information originated
- Memory Profiling: Separate profiles for different contexts
Unlike static chatbots, Ignis develops personality through interactions:
- Curiosity: Learns about user interests and adapts responses
- Wit & Sarcasm: Develops humor based on conversation context
- Empathy: Responds to emotional cues and user mood
- Intelligence: Improves problem-solving through experience
- Context-Aware Tone: Adjusts formality based on situation
- Multi-Modal Responses: Creative, analytical, coding, and professional modes
- Emotional Intelligence: Recognizes and responds to user emotions
- Personality Evolution: Traits change based on interaction history
- Hermes-2-Pro-Mistral-7B: Tuned Mistral model for reasoning and conversation
- GPU Acceleration: CUDA support for faster processing when available
- Model Caching: Reduces load times for better responsiveness
- Async Processing: Non-blocking operations for smooth experience
- Lazy Loading: Components load on demand to reduce startup time
- Modular Design: Organized subsystems for maintainability
- Error Handling: Graceful failure recovery
- Performance Monitoring: Basic metrics and tracking
- iMessage-Style Design: Modern chat bubbles with smooth animations
- Ignis Red Theme: Professional branding with custom color scheme
- Real-Time Updates: Live typing indicators and instant responses
- Memory Dashboard: Visual memory status and conversation statistics
- Rich Commands:
/memory status,/pause learning,/clear memory - Document Processing: Upload and analyze documents
- Code Assistance: Basic coding help
- Calculator: Mathematical computations
Ignis is designed for unlimited expansion:
- Coding Assistant: Python/JavaScript help, code review, debugging
- Document Reader: PDF/text processing with intelligent summarization
- Calculator: Advanced mathematical computations
- Web Search: Local knowledge base queries (expandable to web)
- Async Processing: Non-blocking plugin execution
- Hook System: Pre/post-processing of messages and responses
- Configuration: Per-plugin settings and user preferences
- Security: Sandboxed execution for safe plugin operation
User Input → Command Detection → Memory Retrieval → Context Building → Inference → Response Processing → Memory Storage
- Natural language understanding with intent recognition
- Special command detection (
/help,/memory,/status) - Multi-modal input handling (text, documents, code)
- Vector similarity search across conversation history
- Episodic memory recall with temporal weighting
- Personality trait and emotional state retrieval
- Confidence scoring and relevance ranking
- System prompt integration with personality traits
- Conversation history assembly with memory injection
- Emotional context and mood state incorporation
- Model-specific prompt formatting and optimization
- Hermes-2-Pro-Mistral-7B model processing via llama.cpp
- Dynamic parameter adjustment based on context
- GPU acceleration when available (CUDA/cuBLAS)
- Response generation with quality filtering
- Personality filtering and trait application
- Plugin processing and enhancement
- Emotional response modulation
- Output formatting and presentation
- Atomic fact extraction from conversation
- Vector embedding generation and storage
- Memory consolidation and interference resolution
- Long-term retention with spaced repetition scheduling
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ Episodic │ │ Semantic │ │ Working │
│ Memory │ │ Memory │ │ Memory │
│ (Events) │ │ (Facts) │ │ (Context) │
└─────────────────┘ └─────────────────┘ └─────────────────┘
│ │ │
└────────────────────────┼────────────────────────┘
│
┌─────────────▼─────────────┐
│ Memory Controller │
│ (ChromaDB + SimpleMem) │
│ │
│ ┌─────────────────────┐ │
│ │ Ebbinghaus Curves │ │
│ └─────────────────────┘ │
│ ┌─────────────────────┐ │
│ │ Spaced Repetition │ │
│ └─────────────────────┘ │
│ ┌─────────────────────┐ │
│ │ Interference │ │
│ │ Modeling │ │
│ └─────────────────────┘ │
└───────────────────────────┘
Conversation Input → Sentiment Analysis → Trait Update → Response Filtering → Personality Storage
- Trait Evolution: Personality develops through repeated interactions
- Emotional Learning: AI learns appropriate emotional responses
- Communication Adaptation: Tone and style evolve based on user preferences
- Memory Integration: Personality influences memory formation and retrieval
Ignis/
├── main.py # CLI/Web/API interfaces launcher
├── server.py # FastAPI web server (primary launcher via Ignis.bat)
├── Ignis.bat # **Main launcher** - Double-click to start the AI
├── configs/ # Configuration files
│ ├── conversation_state.json
│ ├── generation_params.json
│ ├── goals.json
│ ├── ignis_profile.json
│ ├── memory_config.json
│ ├── personality.json
│ ├── system_prompt.txt
│ ├── user_config.json
│ ├── user_profiles.json
│ ├── personas/
│ └── prompts/
├── src/ # Modular source code architecture
│ ├── core/ # Essential AI components
│ │ ├── config.py # Configuration management
│ │ ├── context_manager.py # Conversation context building
│ │ ├── emotion_simulator.py # Emotional state simulation
│ │ ├── goals_management.py # Goal tracking and management
│ │ ├── high_performance_inference.py # Optimized inference
│ │ ├── high_performance_memory.py # Memory optimizations
│ │ ├── ignis.py # Main AI logic and personality
│ │ ├── inference_engine.py # llama.cpp inference engine
│ │ ├── memory_system.py # Memory system coordinator
│ │ ├── memory_utils.py # Memory utility functions
│ │ ├── personality_engine.py # Personality processing
│ │ ├── private_config.py # Private configuration (gitignored)
│ │ └── private_goals.py # Private goals (gitignored)
│ ├── memory/ # Advanced memory system
│ │ ├── advanced/ # Cognitive features (spaced repetition, episodic)
│ │ ├── core/ # Essential memory operations
│ │ ├── config/ # Memory configuration
│ │ ├── monitoring/ # Performance tracking
│ │ └── memory_system.py # Memory coordinator
│ ├── interfaces/ # UI implementations
│ ├── plugins/ # Plugin system
│ └── utils/ # Helper utilities
├── memory/ # Memory data storage
├── logs/ # Comprehensive logging
│ ├── application/ # App logs
│ ├── conversations/ # Chat logs
│ ├── memory/ # Memory operation logs
│ └── personality/ # Personality evolution logs
├── data/ # Data storage
│ ├── conversations/ # Chat logs
│ ├── knowledge_base/ # Document storage
│ └── training/ # Learning data
├── models/ # AI model files
│ ├── gguf/ # GGUF model files (Hermes-2-Pro-Mistral-7B.Q4_K_M.gguf)
│ └── mistral/ # Model variants and downloads
├── web/ # Static web files
│ ├── html/
│ │ └── chat.html # Main chat interface
│ ├── css/
│ │ └── chat.css # Styling (iMessage theme)
│ └── script/
│ └── chat.js # Frontend JavaScript
├── scripts/ # Setup and utility scripts
├── tests/ # Comprehensive test suite
├── docs/ # Documentation
├── llama.cpp/ # llama.cpp source code
├── images/ # UI assets
│ └── Ignis.png # Anime character image
└── requirements.txt # Python dependencies
- Double-click
Ignis.bat - Wait for the server to start (takes ~20-45 seconds)
- Chat interface opens automatically in your browser
- Start chatting with Ignis AI!
Ignis will remember everything and evolve its personality through your conversations!
"Ignis memory status" # View memory statistics
"Ignis pause learning" # Temporarily stop learning
"Ignis resume learning" # Resume learning
"Ignis clear memory" # Permanently clear all memories
"Ignis what do you remember about me?"
"Ignis recall our conversation about programming"
"Ignis show my profile"
"Ignis be more sarcastic"
"Ignis tell me a joke"
"Ignis analyze this code"
"Ignis help me debug this"
"Ignis read this file: path/to/document.txt"
"Ignis calculate: (2 + 3) * 4"
"Ignis help me with Python code"
{
"memory_hierarchy": {
"episodic": {
"max_events": 1000,
"retention_days": 365
},
"semantic": {
"vector_dimension": 384,
"similarity_threshold": 0.7
},
"working": {
"max_context": 2048,
"compression_ratio": 0.8
}
},
"cognitive_features": {
"ebbinghaus_enabled": true,
"spaced_repetition": true,
"interference_modeling": true
}
}{
"core_traits": {
"curiosity": 0.9,
"wit": 0.8,
"empathy": 0.7,
"intelligence": 0.95
},
"communication_styles": {
"sarcastic": true,
"metaphorical": true,
"direct": false,
"verbose": "balanced"
},
"evolution": {
"learning_rate": 0.1,
"adaptation_speed": 0.05
}
}{
"default": {
"temperature": 0.85,
"top_p": 0.92,
"max_tokens": 1024,
"repetition_penalty": 1.15,
"gpu_layers": 35
},
"creative": {
"temperature": 1.2,
"top_p": 0.95
},
"analytical": {
"temperature": 0.3,
"top_p": 0.8
}
}- Response Time: 20-60 seconds (CPU), 5-15 seconds (with GPU acceleration)
- Memory Usage: 6-8GB RAM during operation
- Model: Hermes-2-Pro-Mistral-7B.Q4_K_M (7B parameters, 4-bit quantized)
- Startup Time: ~30-60 seconds initial load
- Storage Growth: ~1MB per 100 conversations
- Model Caching: Reduces reload times
- Async Processing: Non-blocking operations
- Lazy Loading: Components load on demand
- Memory Management: Efficient storage
- GPU Support: CUDA acceleration when available
- Context Window: 1024-4096 tokens (configurable)
- Parameter Tuning: Balanced generation settings
- Threading: Concurrent processing
- Memory Pooling: Efficient resource usage
- 100% Local: All processing happens on your machine
- No Telemetry: Zero data collection or external communication
- Private Conversations: Chat history stored locally only
- Model Privacy: Hermes-2-Pro-Mistral-7B runs offline with no API calls
- Memory Encryption: Optional encryption for sensitive memories
- Slow Responses: Enable GPU acceleration or reduce model context
- High Memory Usage: Use quantized models or reduce memory retention
- Startup Delays: Model caching is building - subsequent starts will be faster
- Memory Not Working: Check ChromaDB installation and vector dimensions
- Lost Conversations: Backup
data/directory regularly - Memory Conflicts: Clear memory and restart for fresh initialization
- Model Not Loading: Ensure GGUF file is in
models/gguf/directory - CUDA Errors: Install CUDA toolkit or disable GPU layers
- Out of Memory: Reduce GPU layers or use CPU-only mode
- Clone the repository
- Install dependencies:
pip install -r requirements.txt - Run tests:
python -m pytest tests/ - Start development server:
python server.py --debug
- Memory Features: Add to
src/memory/advanced/ - Plugins: Create in
src/plugins/ - UI Changes: Modify
web/files - Core Features: Extend
src/core/
- Base Model: Mistral-7B-v0.1 with Hermes-2-Pro fine-tuning
- Format: GGUF (optimized for llama.cpp)
- Quantization: Q4_K_M (4-bit mixed precision)
- Context Window: 4096 tokens (configurable)
- Parameters: 7 billion
- File:
models/gguf/Hermes-2-Pro-Mistral-7B.Q4_K_M.gguf
- Storage: JSON-based with atomic facts extraction
- Search: Semantic similarity with confidence scoring
- Backend: File-based storage with indexing
- Features: Episodic memory, spaced repetition, interference modeling
- Backend: FastAPI (ASGI) with async support
- Frontend: Vanilla HTML/CSS/JavaScript
- Database: ChromaDB for vectors, JSON for configuration
- Logging: Structured logging with rotation
- Testing: pytest with async support
- Nous Research for the Hermes-2-Pro-Mistral-7B model
- llama.cpp for efficient local inference engine
- FastAPI for the web framework
- Hugging Face for model hosting and tools
This project is open source under the MIT License. See LICENSE file for details.
Built with ❤️ for local AI
Experience AI that runs privately on your machine.