My Personal AI Agent for Strava

Building a fitness agent to visualize distance, cadence, and calorie burn.

For almost a decade, I’ve been passionate about recording my fitness data. I don’t consider myself a fitness fanatic, but I’ve maintained a routine, from running, crossfit, functional training, or weightlifting at the gym.

For many years, I relied on Google Fit to store this valuable data, accumulating over 10 years of history. However, with the migration (Google health connect) and upcoming deprecation of Google Fit, I was disappointed due to the inaccessibility, although I continue to use Google Fit, Strava, and Huawei Health to review my information, demonstrating my passion for accumulating this type of data, hoping to one day extract value from those 10 years of information.

The agent is “model-agnostic,” functioning with various LLMs like Google Gemini, Llama, or Watsonx. Its architecture is ReAct (Reasoning and Acting), allowing it to reason, select the appropriate Strava tool, and execute to provide accurate responses.

What Your AI Agent Can Do

Designed to leverage custom Strava tools in a conversational interface, the agent enables the discovery of patterns in your training that would otherwise go unnoticed. Based on the repository’s capabilities, the agent covers:

• Granular Activity Analysis: Query recent activities with natural language filter, analyze specific segments/laps, and even break down training zones (heart rate and power).
• Performance Statistics: Calculate total stats by sport, track progress over time, and identify Personal Records (PRs).
• Segment and Route Exploration: Search for segments in specific geographic areas, check club leaderboards, and get detailed elevation and GPS data.
• Advanced Stream Data: For deep analysis, access point-by-point stream data, including GPS, speed, cadence, wattage, and gradient, offering a comprehensive view of every second of your session.

Why I Built It

The main goal was personal improvement through personalization. It’s been very rewarding to review my history, find my longest run, fastest session, or identify performance trends that correlate with the time of year. This transforms the raw data into a narrative about my fitness journey.

If you’re a developer or a fitness data enthusiast, I invite you to explore the code. You can create your own agent and start conversing with your Strava history today.

Note: If you’re an expert in the Google Fit or Health Connect API, contact me. I want to work with over a decade’s worth of data.

Check out the project on GitHub: https://github.com/BrUn3y/Strava_Agent

BeeAI Framework & AgentStack: https://beeai.dev/

Follow me in Strava: https://www.strava.com/athletes/113376809