Everything you need to know about setting up, using, and troubleshooting PancreasAI.
PancreasAI is a personal Android app for people using a Dexcom Continuous Glucose Monitor (CGM). It connects to your Dexcom account to display real-time glucose readings, lets you log insulin doses and meals, and analyzes your data to surface patterns — things like how your glucose responds after meals, how effective your insulin is, and whether you experience the dawn phenomenon.
It also includes optional AI-powered suggestions using Claude, and projected high/low glucose notifications based on your current trend.
You need:
An Anthropic API key is optional — it's only needed if you want the AI Insights suggestions. All other features work without it.
PancreasAI supports two ways to connect to Dexcom:
Dexcom Share (recommended for most users) — uses your standard Dexcom username and password. No developer account required. Go to Settings → Auth Method → Dexcom Share.
Developer API (OAuth) — uses Dexcom's official developer API. Requires a free Dexcom developer account at developer.dexcom.com. Provides a more reliable connection but involves more setup.
To fix it, set a native Dexcom password:
Alternatively, you can use Developer API (OAuth) mode which fully supports Google SSO accounts.
In the Dexcom app on your phone:
PancreasAI uses your own SHARER credentials — enter the credentials of the person wearing the sensor, not a follower account.
Yes. In Settings, toggle Outside US to ON. This switches the Share API connection from the US server (share2.dexcom.com) to the international server (shareous1.dexcom.com). Make sure to save your credentials after changing this setting.
The main screen shows:
Current reading — your latest glucose value in mg/dL, color-coded: ● IN RANGE ● BORDERLINE ● HIGH / LOW
Trend arrow — direction and speed of glucose movement (↑ rising quickly, → stable, ↓ falling, etc.)
Glucose chart — an interactive line graph with your history for the selected time window, with dashed reference lines at your configured high and low thresholds, and markers showing your logged insulin doses and meals.
Stats bar — average glucose, time-in-range %, time low %, and time high % for the current window.
The app refreshes automatically in the background on a configurable interval. The default is every 5 minutes, which matches the Dexcom sensor's reading frequency. You can change this to anywhere between 1 and 30 minutes in Settings → Refresh Interval.
You can also tap the refresh button (↻) on the main screen to pull fresh data immediately.
The chart time window is adjustable using the buttons at the top of the main screen: 6h, 12h, 24h, 3d, 7d, 14d, and 30d.
The app stores up to 13 months of glucose readings locally on your device, so historical data is available even if you've been using the app for a while. The Insights analysis draws from this full local history.
Yes. Go to Settings → Glucose Thresholds. You'll find two sliders:
High threshold — adjustable from 120 to 300 mg/dL (default 180)
Low threshold — adjustable from 50 to 100 mg/dL (default 70)
Changes apply instantly across the entire app — the chart reference lines, IN RANGE / HIGH / LOW labels, time-in-range statistics, and notification triggers all use your configured values.
Tap the export icon in the top toolbar. Choose a time period (24 hours up to 90 days) and the app generates a self-contained HTML report containing your glucose chart, statistics, estimated A1c, insulin log, and meal log for that period.
The report can be shared via email, saved to Google Drive, or printed. It opens in any web browser with no internet connection required.
Tap the syringe button (💉) on the main screen. A dialog will appear where you can enter:
Units — supports up to 2 decimal places (e.g. 12.95 u)
Type — Rapid-acting, Long-acting, or Other
Time — defaults to now, but can be set to any past time
Note — optional free-text field
Doses appear as markers on the glucose chart and are used by the Insights analysis. To delete a dose, swipe left on it in the insulin card or tap the ✕ button.
Tap the food button (🍽) on the main screen. You can enter the food name, carbohydrate grams, calories (optional), meal type (Breakfast, Lunch, Dinner, Snack, Drink, or Other), time, and an optional note.
Meals appear as markers on the glucose chart and feed into the post-meal response curves in Insights.
No. Insulin and food entries are stored entirely on your device. You can log doses and meals offline — they'll appear on the chart and in Insights without any network connection. An internet connection is only needed to pull new glucose readings from Dexcom and to use the optional AI Insights feature.
Tap the lightbulb icon (💡) in the toolbar to open Insights. It analyzes all your stored data and shows:
Time-of-day patterns — your average glucose and time-in-range across 6 four-hour slots throughout the day, making it easy to spot which times are most problematic.
Post-meal glucose curves — how much your glucose rises after meals and how fast it comes back down. Shows the average curve plus individual meal overlays.
Post-insulin response curves — how much your glucose drops after a dose and how long it takes. Shows average and individual dose overlays.
ISF estimate — your estimated Insulin Sensitivity Factor (mg/dL drop per unit of rapid insulin), calculated from your logged doses.
ICR estimate — your estimated Insulin-to-Carb Ratio, derived from meal spikes and your ISF.
Pattern detection — automatic flags for common issues like dawn phenomenon, nocturnal lows, slow-peaking meals, and times with consistently poor control.
ISF (Insulin Sensitivity Factor) — the estimated number of mg/dL that one unit of rapid-acting insulin drops your glucose. For example, an ISF of 40 means one unit should drop you ~40 mg/dL. Requires at least 3 rapid insulin doses logged.
ICR (Insulin-to-Carb Ratio) — the estimated number of grams of carbohydrate that one unit of insulin covers. Derived from your average meal spike and your ISF. Requires at least 5 meals logged.
Dawn phenomenon is a natural rise in blood glucose that occurs in the early morning hours (roughly 2–8 AM), caused by hormones like cortisol and growth hormone released in preparation for waking. PancreasAI detects it by looking for a consistent pattern of glucose rising during the overnight time slots across multiple days.
No. All statistical analysis — patterns, charts, ISF/ICR estimates, and detection — works entirely offline with no API key needed.
The optional AI Suggestions section at the bottom of the Insights screen sends an anonymized summary of your statistics to Claude (Anthropic's AI) and gets back personalized written guidance. To enable it, add your Anthropic API key in Settings → AI Insights. No personal or identifying information is sent — only numbers and timestamps.
Time-of-day patterns and glucose curves improve with more data but show results with just a few days of history. The minimum thresholds for calculated estimates are:
ISF estimation — at least 3 rapid-acting insulin doses
ICR estimation — at least 5 logged meals plus a valid ISF
Pattern detection (dawn phenomenon, nocturnal lows, etc.) becomes more reliable after 7–14 days of combined glucose and logging data.
After every data refresh, the app calculates a weighted rate of change using your most recent 3–4 glucose readings. It then projects that trend forward by your configured projection window (10–40 minutes, default 20) and checks whether the projected value would cross your high or low threshold.
If it would, a notification is fired — for example: "Glucose is 162 and rising. Projected to reach 185 mg/dL in 20 min."
The most recent interval is weighted 2:1 over the prior interval, so sudden changes in direction are accounted for more quickly.
Each alert type (high and low) has a separate 30-minute cooldown. Even if your glucose remains in an alert condition across multiple refresh cycles, you won't receive another notification of the same type within 30 minutes. This prevents repeated pinging during prolonged highs or lows.
High and low alerts are independent — you can receive a low alert within minutes of a high alert if your glucose swings rapidly.
Your high and low thresholds (set in the Glucose Thresholds card above) are what the alerts compare against.
Yes. All health data stored locally uses AES-256-GCM encryption via the Android Keystore — the same standard used by banking and healthcare applications. This includes:
✓ Glucose reading history (up to 13 months)
✓ Insulin dose log
✓ Food / meal log
✓ Dexcom credentials and OAuth tokens
✓ Personal info and app settings
The encryption key lives exclusively in the Android hardware-backed Keystore and never touches the filesystem. This aligns with HIPAA Technical Safeguard requirements for encryption at rest.
Only in these two specific cases:
Dexcom API calls — the app sends your credentials to Dexcom's servers to retrieve your glucose readings. This is the same connection Dexcom's own apps make.
AI Insights (optional) — if you enable this feature with an Anthropic API key, an anonymized statistical summary is sent to Anthropic's Claude API. This contains only numbers (averages, percentages, estimated ISF/ICR) and relative timestamps. Your name, email, Dexcom credentials, and raw glucose timestamps are never included.
All other data — stored readings, insulin logs, food logs, settings — never leave your device.
PancreasAI implements HIPAA Technical Safeguard best practices — specifically encryption at rest (AES-256-GCM), encryption in transit (HTTPS/TLS for all API calls), and minimal data transmission. The app is a personal project, not a formally certified HIPAA Business Associate or covered entity.
This means Dexcom found your account by email but rejected the password for the Share API. The two most common causes:
1. Google SSO account — if your username is a Gmail address and you originally signed up via "Sign in with Google," see the setup question above about Google accounts. You need to set a native Dexcom password via the Forgot Password flow.
2. Wrong password — verify you can log into dexcom.com with the credentials stored in the app. If you recently changed your Dexcom password, update it in Settings.
Check the following:
Go to Settings and tap Run Diagnostics. The tool walks through every login method step by step and displays the raw HTTP responses from Dexcom's API. It shows:
• Which login flow succeeded or failed and why
• The exact error code Dexcom returned (e.g. AccountPasswordInvalid)
• How many readings were returned and what the latest one is
• Specific warnings like the Google SSO issue flagged inline
This output is useful to share when asking for support.
Time-of-day patterns and glucose curves appear with just a few days of glucose history. For the calculated metrics:
ISF — log at least 3 rapid-acting insulin doses
ICR — log at least 5 meals (and you'll need ISF first)
The more data you log, the more accurate the estimates become. Pattern detection is most reliable after 1–2 weeks of combined glucose and logging data.
This should not happen. The app automatically migrates plain-text data from older versions to the encrypted format on first launch — your data should carry over seamlessly.
If data does appear missing, it may be because Android cleared the app's data storage during a reinstall rather than an upgrade. Check Android Settings → Apps → PancreasAI → Storage to see if data was cleared. Unfortunately, data cleared by Android's storage manager cannot be recovered.