From zero to a working GlycemicGPT setup -- platform, mobile app, and (optionally) the watch face.
This guide walks you through setting up GlycemicGPT end-to-end: the platform itself, the Android companion app you need to connect your pump, and the optional watch face.
Try it locally -- run on your own laptop or desktop
Plan to use it day-to-day with your phone reaching it from anywhere
Always-on deployment -- a home server, mini-PC, NAS, Raspberry Pi, or a small cloud server you rent (sometimes called a VPS)
You can always start locally and migrate to an always-on setup later -- your settings and data live in the database, and the platform itself is the same.
Before you start, you need:
A computer to run the platform on -- a laptop, desktop, home server, or rented cloud server. The platform is light:
RAM: 2 GB minimum, 4 GB recommended. (One part of the platform uses about 1 GB on its own to do the AI knowledge-base lookup; the rest is light.)
Disk: 5 GB minimum, 10 GB recommended. The platform's files take about 3 GB; the rest grows slowly with your data over time.
CPU: any modern dual-core works. The platform sits mostly idle except when the AI is answering a question.
OS: macOS, Linux, or Windows. Windows users: see the note below before you start -- you'll need to install one extra Microsoft component first.
An Android phone for the companion app (required to connect your pump over Bluetooth)
Windows users only: GlycemicGPT runs on Linux software. Windows lets you run that software through a Microsoft feature called WSL2 (Windows Subsystem for Linux 2). It's free and Microsoft-supported, but it does add some setup steps before this guide. Plan to spend roughly an extra half-hour up front following Microsoft's WSL2 setup guide. After that, the rest of this guide works the same as it does on macOS or Linux. macOS and Linux users can skip this note.
Several steps below ask you to "run a command in the terminal." If you've never used a terminal before, that's fine -- here's the short version:
The terminal is a text-based way to control your computer. It's already installed on every Mac, Linux machine, and Windows computer. You type commands; the computer runs them.
To open it:
macOS: open Terminal (it's in Applications → Utilities, or press Cmd+Space and type "Terminal")
Linux: open Terminal from your applications menu (the exact name varies by distribution -- Terminal, Console, GNOME Terminal, Konsole)
Windows: open Ubuntu (or whatever WSL2 distribution you installed; see step 1 if you don't have WSL2 yet)
When this guide shows a code block like:
some-command
...it means: type that line into the terminal and press Enter.
"Copy-paste" works in the terminal -- you don't have to type long commands by hand. Right-click and paste, or Cmd/Ctrl+Shift+V depending on your terminal.
You'll keep the same terminal window open through most of this guide. Don't close it until you're done.
git clone ... -- downloads all the GlycemicGPT files onto your computer. It creates a folder named GlycemicGPT wherever you were when you ran the command. (If you don't have git installed, your terminal will say so. To install it: xcode-select --install on Mac, sudo apt install git on Linux, or download from git-scm.com on Windows.)
cd GlycemicGPT -- "change directory" -- moves you into the folder you just downloaded. From here on, every command in this guide should be run from inside this folder unless we say otherwise.
Each example folder has a README.md with the start-to-finish walkthrough for that path. Below is a quick reference for the three most common cases. If you're not sure which to pick, the home-server-with-Cloudflare-Tunnel path is the easiest "I want to actually use this day-to-day" option for most users, and the laptop path is the easiest "I want to try it" option.
In the same terminal window, still inside the GlycemicGPT folder, run:
docker compose up -d
What this does: docker compose up reads the recipe in docker-compose.yml and starts all the parts GlycemicGPT needs:
a database that stores your data
a cache that speeds up repeat requests
the API that the dashboard and the mobile app talk to
the web dashboard itself (what you'll open in a browser)
the AI bridge that forwards your AI chat messages to whichever AI provider you set up later
The -d flag means "run them in the background" so they don't take over your terminal window.
That's it. The platform is running on your computer at http://localhost:3000.
The first time you run this, it will take a few minutes to download images and build everything (you'll see a lot of text scrolling by -- that's normal). Subsequent starts are fast. There's no public exposure -- the dashboard is only reachable from your computer.
This is the path most users will want for day-to-day use. You run GlycemicGPT on either:
a computer at home (desktop, NAS, mini-PC, Raspberry Pi -- anything running 24/7), or
a cloud VPS without opening any inbound ports to the internet.
Cloudflare manages the public access. No port forwarding, no public IP from your ISP, no inbound firewall rules on your VPS, no TLS certificates to renew. Your server makes one outbound connection to Cloudflare; all inbound traffic comes through that connection. This is often the simplest path for both home and VPS deployments and is generally more secure than opening ports directly to the internet -- see Install with Docker -- Why this is often more secure than opening ports for the security rationale.
What you'll need: a Cloudflare account (free) and a domain on Cloudflare.
cd deploy/examples/cloudflare-tunnel/cp .env.example .env# Edit .env -- paste your Cloudflare Tunnel token + generate the secretsdocker compose up -d
The full walkthrough -- creating the Cloudflare account, adding your domain, creating the tunnel, configuring routing -- is in Install with Docker -- Deploying with Cloudflare Tunnel. It's written for non-technical users with no prior Cloudflare experience.
This is the path for users who don't have a home server (or don't want to run one) and would rather rent a small cloud server. You get a domain pointing at the server's public IP, Caddy provisions HTTPS automatically via Let's Encrypt.
What you'll need: a VPS from any provider (Hetzner, DigitalOcean, Linode, AWS Lightsail, etc.) and a domain you control.
cd deploy/examples/public-cloud/cp .env.example .env# Edit .env -- set DOMAIN, ACME_EMAIL, and generate the secretsdocker compose up -d
The full walkthrough -- DNS setup, firewall, certificate provisioning, troubleshooting -- is in:
docker compose ps shows the status of each service. You're ready when each service shows healthy or running. If anything still shows starting, wait another 30 seconds and run the command again to check.
Always-on deployment: the public hostname you configured in step 4 (for example https://glycemicgpt.yourdomain.com for a Cloudflare Tunnel subdomain, or https://yourdomain.com for a VPS root domain)
Most people land here. Read the row that matches you, then jump to that option below:
If this is you...
Do this
Cost
You already pay for Claude (Pro / Max)
Option 1
$0 extra (uses your subscription)
You already pay for ChatGPT (Plus / Team)
Option 2
$0 extra (uses your subscription)
No subscription, want the easiest cloud option
Option 3 (Claude API key)
A few cents to a few dollars per month
No subscription, want the cheapest cloud option
Option 4 (OpenAI API key, cheaper models)
Often under $1 per month
You want everything to stay on your own computer
Option 5 (local Ollama)
$0, but you need a beefy enough computer
If you're not sure which to pick, the strongest recommendation for typical users is Option 1 if you already pay for Claude, otherwise Option 3. You can switch at any time from Settings → AI Provider without losing data.
For a deeper comparison (privacy, quality, switching), see BYOAI.
If you already pay for Claude (Pro, Max, or Team), you can use that subscription with GlycemicGPT -- you don't need a separate API key. The platform's AI bridge uses a one-time login token from Anthropic's official command-line tool to make calls under your subscription.
You'll need Node.js installed on a computer (your laptop is fine -- it doesn't have to be the same computer the platform runs on). If you don't have Node.js: download the LTS version from nodejs.org and run the installer.
Once Node.js is installed, open a terminal and run:
npx @anthropic-ai/claude-code setup-token
This opens a browser window where you sign in to your Claude account. After you sign in, the terminal prints a long string of letters and numbers -- that's the token. Copy it.
In the GlycemicGPT dashboard, go to Settings → AI Provider, choose Claude (subscription), and paste the token. The AI bridge stores it and uses it for all your AI requests from now on.
Heads up on subscription tokens: The CLI commands above (claude-code setup-token, codex login) are official, vendor-published tools, but using their tokens from a third-party server (the AI bridge) for ongoing background calls is not a use case Anthropic or OpenAI has publicly endorsed. It works today, but tokens can stop working without notice if a vendor update changes how they're scoped. If you need rock-solid reliability (caregiver alerts, someone else's data flowing through this), use Option 3 or 4 -- direct API keys are explicitly designed for server use. See BYOAI -- Subscription-token note for the longer discussion.
If you already pay for ChatGPT (Plus, Team, or Enterprise), you can use that subscription with GlycemicGPT. Same idea as Claude -- you'll use OpenAI's official command-line tool to get a token.
You'll need Node.js installed (see Option 1 if you don't have it yet). Then run in a terminal:
npx @openai/codex login
A browser window opens for you to sign in to your OpenAI account. The terminal then prints a token. Copy it, then paste it into the GlycemicGPT dashboard at Settings → AI Provider → ChatGPT (subscription).
You can change providers anytime in Settings → AI Provider. Your data stays where it is; the platform just starts routing AI calls to the new provider.
The Android app is required to connect your pump and read its data. The platform alone cannot do this -- the phone app handles the Bluetooth connection.
See Mobile app install for the step-by-step Android install (about 10-15 minutes). The short version:
On your phone, open the file and install it (you'll need to allow installs from unknown sources)
Open the app, point it at your platform's URL (local: your computer's IP; always-on deployment: your domain), and sign in with the account you just created
Once the app is signed in and your pump is paired, your dashboard fills with data.
A Wear OS watch face is available for at-a-glance glucose viewing. It's optional -- the platform and phone app give you everything important without it.
Heads up: setting up the watch face is more involved than the rest of this guide. It requires connecting your watch to a computer over a developer connection (called ADB) to install the watch face directly, since the watch face is not on the Play Store yet.
GlycemicGPT is alpha software. It works, but it has not been broadly tested. Treat it as a tool that supplements your existing diabetes management, not as a replacement for it.
GlycemicGPT does not control insulin delivery. It reads from your devices; it does not write back to them.
AI suggestions are not medical advice. When the AI tells you something interesting, talk to your endocrinologist before acting on it.
Your data stays with you. GlycemicGPT does not phone home or send your data anywhere. The AI provider you choose to use will see the messages you send through the chat interface, just like any other AI service.