CalmLoop — Responsible AI Wellness Agent

Overview

CalmLoop is an early-stage AI wellness companion designed to support daily stress, energy, and recovery through short, adaptive check-ins. The project focuses on responsible agent design, safety boundaries, and observability rather than feature completeness.

Current Status

This repository represents a checkpoint submission. The core agent logic, safety constraints, and observability design are implemented conceptually, with a minimal executable scaffold.

Intended Functionality

• Users provide short daily self-reports (mood, energy, stress, sleep).
• A constrained LLM-based agent reasons over the inputs.
• The agent selects one low-risk wellness suggestion.
• The agent explains why the suggestion fits the current context.

Agent Design

• Single responsibility-bounded agent
• No diagnosis, prediction, or medical advice
• Structured reasoning: Input → Assessment → Recommendation → Explanation

Safety Principles

• Not a medical device
• Avoids clinical language and prescriptive claims
• Encourages user autonomy and appropriate professional support

Evaluation & Observability (Planned)

The system is designed to integrate with Comet / Opik for LLM observability:

• Logging of agent inputs
• Logging of reasoning summaries
• Logging of outputs
• Optional user feedback signals

This enables inspection, comparison of prompt iterations, and future human-in-the-loop evaluation.

Scope (Intentional)

Included:

• Core agent logic
• Safety constraints
• Observability design

Excluded:

• Wearables or biometric data
• Long-term prediction
• Multi-agent systems
• Clinical claims

Notes

This checkpoint emphasizes responsible system design and evaluability. Feature expansion and deployment are planned for later stages.

Run locally

python3 main.py