Your Personal AI That Learns and Grows With You
GroundZero AI is a complete AI system that continuously learns from your interactions, builds a knowledge graph, remembers conversations, and improves over time. It's built to be "your own" AI that you can train and customize.
- Chain-of-thought reasoning with visible thinking steps
- Self-verification against knowledge graph and web
- Confidence scoring for all responses
- Stores and connects concepts, facts, and relationships
- Grows automatically as you interact
- Verifies information from multiple sources
- Semantic search and path finding
- Conversation history - remembers all chats
- User profiles - learns your name, preferences, behaviors
- Long-term memory - stores facts and events
- Working memory - context for current conversation
- Multi-engine search (DuckDuckGo, Wikipedia, arXiv)
- Source reliability scoring
- Fact verification with evidence
- Deep research mode for comprehensive learning
- Learns from your feedback (👍/👎)
- Learns from corrections ("no, that's wrong...")
- Learns topics from the web
- Background learning cycles
- Run Python code with full output capture
- Execute bash/shell commands
- Install packages on the fly
- Persistent execution environment
- Read ANY file type (PDF, Excel, Word, CSV, images, etc.)
- Extract tables, text, and structure
- Ask questions about document content
- Multi-document analysis for analytics
- Create Word documents (.docx)
- Create PDF files
- Create Excel spreadsheets (.xlsx)
- Create PowerPoint presentations (.pptx)
- Create CSV, Markdown, and more
- Beautiful chat interface
- Real-time reasoning display
- Conversation history
- Knowledge graph visualization
- User settings
cd groundzero
pip install -r requirements.txtpython test.pypython run.py
# or
python run.py --chatpython run.py --dashboardThen open http://localhost:8080
groundzero/
├── config/
│ └── config.yaml # Main configuration
├── data/ # All data storage
│ ├── knowledge/ # Knowledge graph
│ ├── memory/ # Long-term memory
│ ├── models/ # Downloaded models
│ ├── training/ # Training data
│ ├── conversations/ # Chat history
│ └── users/ # User profiles
├── src/
│ ├── core/ # Model management
│ ├── knowledge/ # Knowledge graph system
│ ├── memory/ # Memory system
│ ├── search/ # Web search & verification
│ ├── reasoning/ # Chain-of-thought reasoning
│ ├── continuous_learning/ # Learning from interactions
│ ├── dashboard/ # Web interface
│ │ ├── templates/ # HTML
│ │ └── static/ # CSS & JS
│ ├── utils/ # Utilities
│ └── groundzero.py # Main AI class
├── run.py # Entry point
├── test.py # Test suite
└── requirements.txt
from src.groundzero import GroundZeroAI
ai = GroundZeroAI()
# Chat
response, reasoning = ai.chat("Hello! What can you do?", return_reasoning=True)
print(response)
# Show reasoning
if reasoning:
print(f"Confidence: {reasoning.confidence:.0%}")
for step in reasoning.steps:
print(f" {step.step_number}. {step.thought}")# Teach directly
ai.teach("Python", "Python is a programming language created by Guido van Rossum")
# Have it learn from the web
ai.learn("machine learning transformers")# Rate a response
ai.feedback("What is X?", "Response...", rating=5) # 1-5
# Correct a mistake
ai.correct(
prompt="What is the capital of Australia?",
wrong_response="Sydney",
correct_response="Canberra"
)# Search knowledge
result = ai.ask_knowledge("machine learning")
print(result["results"])
# Verify a fact
verification = ai.verify_fact("The Earth orbits the Sun")
print(f"Verified: {verification['verified']}, Confidence: {verification['confidence']}")Edit config/config.yaml:
model:
name: "GroundZero-AI"
version: "1.0.0"
base_model: "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B"
quantization: "4bit" # For GPU memory efficiency
knowledge:
auto_verify: true
min_confidence: 0.7
search:
engines:
- duckduckgo
- wikipedia
- arxiv
verify_sources: true
continuous_learning:
enabled: true
auto_evolve_threshold: 100
dashboard:
host: "0.0.0.0"
port: 8080- Python 3.9+
- 4GB RAM
- Works on any CPU
- Python 3.9+
- 16GB+ RAM
- NVIDIA GPU with 8GB+ VRAM (RTX 3060+)
- CUDA 11.8+
User Input → Generate Response → Observe Interaction
↓
[Learning Signal Queue]
↓
Feedback (👍/👎) ─────────────────→ Process Signals
Corrections ──────────────────────→ ↓
Web Research ─────────────────────→ Training Data
↓
[Training Cycle]
↓
Improved Model
- Observe: Every interaction is recorded
- Feedback: Ratings create learning signals
- Corrections: Direct corrections are high-priority
- Research: Web learning adds knowledge
- Evolve: Training on accumulated signals
- Clean, modern design
- Typing indicators
- Reasoning panel
- Feedback buttons
- Browse past conversations
- Search chats
- Visualize the knowledge graph
- Search concepts
- Model status
- Knowledge count
- Learning progress
- All data stored locally in
/data - No external transmission except explicit web searches
- User profiles stay on your machine
- You control what it learns
# Interactive chat
python run.py --chat
# Web dashboard
python run.py --dashboard --port 8080
# Learn about a topic
python run.py --learn "quantum computing"
# Download model (requires GPU)
python run.py --download| Method | Description |
|---|---|
chat(message) |
Chat and get response with optional reasoning |
learn(topic) |
Research and learn from web |
teach(subject, content) |
Directly add knowledge |
feedback(prompt, response, rating) |
Rate a response (1-5) |
correct(prompt, wrong, correct) |
Correct a mistake |
ask_knowledge(query) |
Query knowledge graph |
verify_fact(claim) |
Verify a fact |
evolve() |
Run training on learning queue |
get_stats() |
Get system statistics |
save() |
Save all data |
Areas to help:
- Better reasoning algorithms
- More search engines
- UI improvements
- Test coverage
- Documentation
MIT License - Use freely!
- Built on DeepSeek-R1-Distill
- Inspired by Claude AI's interface
- Uses QLoRA for efficient fine-tuning
Made with ❤️ for AI enthusiasts who want their own personal AI