Open-source AI assistant for diabetes management and monitoring

Who: Posted by jlengelbrecht on Hacker News, sharing their own project: GlycemicGPT, an open-source diabetes management platform. The developer describes daily personal use, suggesting jlengelbrecht is themselves a person with diabetes building tooling for their own care.

What's new: GlycemicGPT is a self-hosted, open-source platform that pulls real-time data from a CGM and an insulin pump, then feeds that data to an AI provider of the user's choice to generate daily summaries, detect patterns, and answer questions via chat. It is currently in alpha, supports the Dexcom G7 CGM and Tandem t:slim X2 pump via both BLE and cloud APIs, and can layer on top of an existing Nightscout installation without disturbing it.

How it works: The system runs as a self-hosted Docker stack composed of several services: a Next.js frontend, a FastAPI backend, a PostgreSQL database, a Redis cache, and an AI sidecar that acts as a proxy to whichever AI provider the user supplies — OpenAI, Anthropic Claude, or a locally-run model via Ollama. A native Android app with a Wear OS companion handles mobile monitoring and direct pump communication. A plugin architecture allows community contributors to add support for additional devices.

Caveats: The project ships with prominent disclaimers: it is not approved by the FDA or any regulatory body, and the developer explicitly warns that AI models can hallucinate or misinterpret glucose data. Support for devices beyond the Dexcom G7 and Tandem pumps is unverified or still planned, and the Nightscout integration is listed as forthcoming rather than live. The alpha label is earnest: this is one developer's personal tooling made public, not a clinically validated product.