Home

GlucoPilot Wiki

A self-hosted, single-user personal health platform centered on Type 1 diabetes. It unifies glucose, insulin, wearables, cycle, lab work, and a daily symptom journal into one private app — then lets you talk to all of it. The centerpiece is a health Companion: a chat grounded in your real data that remembers your lived experience and reasons across every domain. Everything runs on your server with your keys.

⚕️ Not a medical device. GlucoPilot is for personal data exploration and education. It does not diagnose, recommend insulin dosing, or control any device. Always consult your care team for treatment decisions.

Start here

• Getting Started — run it locally in one command, or deploy on a server.
• The Companion — the health chat that reasons across all your data.
• Connecting Data Sources — Dexcom, Nightscout, Tandem, Glooko, Oura, Google Health, imports.
• AI and Privacy — choose a cloud provider or run everything locally.
• Sharing With Your Doctor — provider logins and the Visit Report.
• FAQ

What it does

Area	Summary
Companion	a chat grounded in your real records — remembers what you tell it, keeps multiple threads, switches between a fast and a deeper local model
Overview	a cross-domain AI summary that spots connections across your whole picture
Glucose	real-time (Dexcom Share) + history; TIR/GMI/CV, AGP, a zoomable Explorer chart
Insulin / pump	boluses, basal, suspends from Tandem Source or Glooko; TDD, estimated resistance, absorption stats
Wearables	Oura (sleep/HRV/temperature) and Fitbit / Google Health (steps/HR/sleep/SpO₂), incl. near-real-time heart rate
Cycle	phases inferred automatically from Oura nightly temperature
Symptom journal	a nightly check-in (severity, duration, notes) woven into the AI and the report
Patterns & Insights	recurring highs/lows, post-meal spikes, dawn phenomenon; cross-domain correlations
Records	upload lab PDFs/photos → extracted and grouped into per-analyte trend charts
Visit Report	printable 90-day clinical summary (conditions, meds, labs, symptoms) + an AI "quarter in review"

Design principles

• One owner per deployment. Single-tenant by design — you're the sole custodian of your data. Run one instance per person; it is not built to host other people's health data.
• Your keys, your server. Integrations use your own API credentials; a local model keeps everything on your machine.
• Honest about unofficial APIs. Some connectors (Dexcom Share, Tandem Source, Glooko) use community-maintained unofficial APIs that can change.