Self-hosted, open-source diabetes platform connecting Dexcom G7 and Tandem pumps to BYOAI analysis (Claude, OpenAI, Ollama); alpha, built by the developer for daily personal use.
Key Takeaways
BYOAI architecture: plug in your own AI provider key; a TypeScript/Express sidecar proxies to any OpenAI-compatible endpoint.
Stack is Next.js 15 + FastAPI + PostgreSQL 16 + Redis 7, deployable via Docker Compose or Kustomize-based Kubernetes with Cloudflare Tunnel examples included.
Android + Wear OS companion app reads Tandem t:slim X2 over BLE directly; Tandem Mobi is protocol-compatible but unverified on physical hardware.
Nightscout integration (coming soon) will extend device coverage: anything feeding Nightscout will feed GlycemicGPT without reconfiguring existing setups.
Does not control medical devices; suggestions only, with a pre-validation layer and caregiver escalation via Telegram, push, and in-app alerts.
Hacker News Comment Review
The single comment questions whether the caregiver alerting adds real value over existing solved solutions like Dexcom Follow and Abbott LibreLinkUp, and whether AI pattern analysis is meaningful without consistent meal logging from the user.
No consensus or technical implementation discussion has emerged yet beyond that skepticism.
Notable Comments
@fnands: flags that alerts and caregiver sharing are already solved by Dexcom Follow and LibreLinkUp, questioning whether the AI analytics delivers value without logged meal data.
