Biological AI Memory — Your thoughts, remembered forever.
HEARTBEAT is a premium, localized AI memory system engineered to function like a living biological brain. It doesn't just respond to your prompts; it metabolizes your conversations into permanent, searchable "Memory Cells" that circulate in real-time. HEARTBEAT provides an everlasting context window, granting you an AI companion that grows, learns, and perfectly remembers every detail about you.
Traditional AI chatbots suffer from global amnesia the moment a session ends. HEARTBEAT permanently solves this by actively extracting concepts, facts, and sentiments from every interaction, routing them through a multi-stage Intelligence Pipeline.
- The L1 Sieve: Rapidly categorizes incoming thoughts and filters out noise.
- The Ambiguity Detector: Clarifies missing context before committing anything to memory.
- The Cell Generator: Synthesizes raw data into immutable
Memory Cells(Vector Embeddings + Metadata). - The Nervous System: Streams live telemetry back to the dashboard, letting you physically watch your AI brain form memories in real-time.
- Real-time Cellular Dashboard: A professional, three-column UI reminiscent of Claude or ChatGPT, but heavily upgraded. Watch your
Active Cellspulsate and circulate in the telemetry panel as the AI thinks. - 3-Level Answer Depth Synthesis: Tailor your AI's recall on-the-fly. Choose between Summary (Quick overview), Detail (Comprehensive synthesis), or Original (Exact retrieval) under every AI response.
- Persistent Neural Storage: Unbounded long-term memory powered by Local Vector Databases (ChromaDB) and Relational Storage (SQLite).
- Clarification Loops: If HEARTBEAT detects an ambiguous pronoun or statement, it triggers a specialized UI banner to ask for clarification, ensuring high-fidelity memories.
- Local-First Design & OCR: Secure privacy limits by processing state and images locally using
pytesseract, while firing rapid logical deductions via Groq's high-speed API.
HEARTBEAT is meticulously engineered for high-performance scale and raw speed.
- Language: Python 3.10+
- API Framework: FastAPI (Serving as the high-speed Artery)
- Real-time Telemetry: WebSockets (Serving as the Nervous System)
- LLM Inference Engine: Groq API (Primary routing for ultra-low latency), OpenRouter (Fallback API)
- Vector Embeddings:
sentence-transformers(Local generation)
- Semantic Memory: ChromaDB (Vector database for similarity searches and RAG mechanics)
- Relational Backbone: SQLite (Pydantic v2 schemas for tracking cell scores, topics, and session history)
- Image Processing:
Pillowandpytesseract(Local OCR pipeline) - NLP Utilities:
nltkfor chunking and lexical analysis
- Structure: Vanilla HTML5
- Styling: Custom CSS3 with CSS Variables, Flexbox/Grid, and Premium Micro-animations
- Logic: Vanilla ES6+ JavaScript,
localStoragestate persistence, native fetch APIs, and rawWebSocketlisteners.
- Framework:
pytestwithunittest.mock
- Python 3.10+
- Tesseract-OCR installed on your system
- A Groq API key (and optionally an OpenRouter key)
Clone the repository and install the biological dependencies:
pip install fastapi uvicorn pydantic python-dotenv chromadb sentence-transformers pillow pytesseract nltk groq httpx pytest redisCreate a .env file in the root directory:
# .env
HEARTBEAT_BRAIN_KEY=your_groq_api_key_here
HEARTBEAT_OPENROUTER_KEY=fallback_key_here-
Start the Artery (Backend):
$env:PYTHONPATH="."; python main.py
The server runs locally on
http://127.0.0.1:8000 -
Launch the Dashboard: Double-click
ui/index.htmlor run it via Microsoft Edge/Chrome. Enjoy a seamless, living UI capable of managing unbounded conversations.
“The real magic of intelligence is not in the generation of text, but in the permanent formation of memories.”