Commenters debated ways to keep Garmin (and other device) data local and usable by personal AIs. Several people described practical solutions: connecting devices over USB and dragging .fit/.gpx files, using GadgetBridge, GoldenCheetah, intervals.icu, or run-gap to ingest activities, or writing agents that pull unofficial Garmin APIs into sqlite/Notion for Claude-style briefs. A few shared code and projects - an “AI skill” for Garmin, a running-coach agent, and various MCP-compatible servers and apps (Coros, SparkyFitness, SparkyFitness, Intervals.icu). Others explained workflows like syncing to Strava then using its API, exporting weekly sheets for LLM analysis, or running small scripts to upload raw files after local sync.
Opinion split over tradeoffs and risks. Some commenters pushed for MCPs/open APIs to enable custom feeds and personal algorithms, while others warned vendors would resist because it eases competition or scraping. Practical concerns were raised about syncing complexity, vendor post-processing (GPS smoothing and elevation corrections) being lost with raw exports, and vendor fees for intermediary services. Privacy and security worries were common: several people said they’d never send sensitive personal data or financial permissions to commercial LLMs, preferring local models for private tagging and analysis.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.