GlycemicGPT: Self-Hosted AI Diabetes Monitor with BYOAI and Plugin SDK
GlycemicGPT is an open-source diabetes management platform built for self-hosting. It connects continuous glucose monitors (CGM) and insulin pumps to an AI analysis layer that runs on your own infrastructure. The project was created by a Type 1 diabetic software engineer who needed a tool to review data between endocrinologist visits.
Supported Devices
- Dexcom G7 – via cloud API (verified)
- Tandem t:slim X2 – direct BLE + cloud API (verified)
- Tandem Mobi – BLE (protocol-compatible, not verified on physical hardware)
- Nightscout – point at your existing instance
AI Layer Features
- Daily briefs summarizing overnight and 24-hour patterns
- Meal response analysis
- Conversational chat backed by RAG with clinical knowledge
- Predictive alerting with configurable thresholds and caregiver escalation
Important: GlycemicGPT does not deliver insulin, control pumps, or function as a closed-loop system. It reads data and provides insights only.
Architecture
- Self-hosted via Docker or Kubernetes – runs entirely on your hardware
- BYOAI – bring your own AI provider: Ollama for fully local operation (zero data leaves your hardware), or any OpenAI-compatible endpoint (Claude, OpenAI, etc.)
- Data flows directly from your instance to the chosen provider; no centralized project-operated services
Stack
- Backend API: FastAPI, Python 3.12, PostgreSQL 16, Redis 7
- Web Dashboard: Next.js 15, React 19, Tailwind CSS, shadcn/ui
- AI Sidecar: TypeScript, Express, multi-provider proxy
- Android App: Kotlin, Jetpack Compose, BLE
- Wear OS: Kotlin, Wear Compose, Watch Face Push API
- Plugin SDK: Kotlin interfaces, capability-based, sandboxed
Quick Start
Deploy with Docker Compose, configure a .env file with your device credentials and AI provider endpoint, then run:docker compose up -d
You can point it at an existing Nightscout instance for immediate data ingestion.
Who It's For
Developers with diabetes who want to self-host their own AI-powered monitoring stack, and contributors interested in BLE/Android or diabetes device integrations.
📖 Read the full source: HN AI Agents
👀 See Also
DESIGN.md: A format spec for describing visual identity to coding agents
DESIGN.md combines YAML design tokens with markdown prose to give AI coding agents a persistent, structured understanding of a design system. Includes a linter and diff tool.
RunAnywhere RCLI: On-Device Voice AI Pipeline for Apple Silicon
RunAnywhere has released RCLI, an open-source voice AI pipeline for macOS that runs STT, LLM, and TTS entirely on Apple Silicon devices. The tool uses their proprietary MetalRT inference engine and claims significant performance improvements over existing solutions.
OpenLobster: Self-Hosted AI Agent in Go with 30MB RAM Footprint
OpenLobster is a self-hosted AI assistant written in Go that runs as a single binary with 30MB RAM usage and 200ms cold start. It supports multiple LLM providers including Ollama, OpenRouter, and any OpenAI-compatible endpoint, with memory stored in a graph database.
DeepClaude swaps Claude Code's Anthropic backend for DeepSeek V4 Pro at 17x lower cost
A script that rewrites Claude Code's environment variables to route all agent loop calls through DeepSeek V4 Pro, OpenRouter, or Fireworks AI — same UX, $0.87/M output tokens vs $15/M.