Harlo
Acts as an AI coach that observes user patterns, predicts cognitive states like burnout, and manages persistent cognitive state using USD composition semantics.
Harlo is your personal AI coach designed to enhance productivity and well-being by intelligently monitoring your work patterns. It predicts potential crashes, intervenes before burnout occurs, and scales back during periods of flow. Uniquely built on USD composition semantics, Harlo ensures persistent, local-first cognitive state management, storing all data on your device as composable USD layers. This architecture guarantees no cloud dependency, no data mining, and complete ownership of your cognitive data, providing a private and robust solution for personal AI.
Key Features
01Local-First Cognitive State Management: Stores all cognitive state locally as time-sampled, human-readable, and Git-trackable USD layers, ensuring data privacy and ownership.
02Graceful Degradation: Designed with independent component failure handling, ensuring the system remains operational even if individual parts like USD import or delegate execution encounter issues.
03Modular LLM Integration with Hydra Delegates: Supports flexible integration of various Large Language Models (LLMs) via a capability-matching delegate pattern, ensuring the right LLM is used for specific tasks (e.g., reasoning, code generation).
04Proactive Burnout Prediction: Utilizes an XGBoost predictor trained on synthetic and organic data to forecast momentum, burnout, energy, and burst, intervening before exhaustion.
054 GitHub stars
06Biologically-Architected Memory System: Employs a multi-tiered memory system including a Rust-based Hippocampus crate for sub-2ms recall, FTS5 for hot-tier storage, and Hebbian learning for SDR evolution.
Use Cases
01Personal productivity enhancement and workflow optimization.
02Proactive prevention of burnout and cognitive overload.
03Tailored AI coaching for improved focus and well-being.