Metabolic health went from a niche to a category in about two years. Over-the-counter continuous glucose monitors put a sensor on people without diabetes, GLP-1 medications created a large population of users who want to see what their body is doing, and a wave of coaching and nutrition products now assume glucose data as an input.
Most founders entering this space assume the first step is "call the Dexcom API." That's usually the wrong first step, and sometimes the wrong step entirely. Here are the four real routes to CGM data, what each costs, and which one we recommend for a first version.
Route 1: read glucose from HealthKit or Health Connect
This is the one nobody writes about and the one most products should use first.
Both Dexcom's and Abbott's consumer apps can write glucose readings into the phone's health store. Which means: if your user already has the manufacturer's app installed and configured — and they do, because that's how the sensor works — you can read their glucose through the same HealthKit or Health Connect integration you'd build for steps and sleep.
What you get: glucose readings, no vendor partnership, no separate OAuth, no rate limits, and no additional approval. On iOS, blood glucose is one of the data types Apple treats as time-critical for background delivery, so you can be notified relatively promptly rather than polling.
What you don't get: anything the manufacturer's app doesn't write. No sensor session metadata, no calibration state, no trend arrows, no manufacturer-computed metrics. And your data is only as good as the user's own sync.
Cost: it's part of a normal $3,500 platform integration, not a separate project.
For a first version of a metabolic-health or GLP-1 companion product, this is usually enough to find out whether anyone uses the feature. Start here. If your product then genuinely needs sensor sessions and trend arrows, go to route 2 having learned something.
Route 2: the Dexcom API
Dexcom runs a real developer program with documentation, a sandbox environment and OAuth 2.0. You can build against the sandbox today.
The thing to understand before you design anything: the general developer API is not real-time. Data is returned on a delay. This is deliberate — Dexcom does not want third-party apps positioning themselves as real-time glucose monitoring or alerting without the appropriate relationship and regulatory posture. Real-time access sits behind a separate partner arrangement.
The design consequences are absolute:
- You cannot build alerting on it. No "your glucose is dropping" notifications. Not "we'll add a disclaimer" — the data is too old to be safe, and building it anyway is the kind of decision that ends a company.
- You cannot build a live dashboard. Present the data as retrospective. "Yesterday's curve" is a legitimate product. "Your glucose right now" is not.
- Retrospective analysis is the product. Time in range, post-meal responses, overnight patterns, correlations with food, sleep and activity. All of these work fine on delayed data.
Dexcom also has an established estimated-glucose-value data model with regular interval readings, event data (meals, exercise, insulin where the user logged it), calibrations and device/sensor session information — richer than what lands in HealthKit.
Route 3: Abbott / FreeStyle Libre
The honest answer is short: Abbott does not offer a broadly open public developer API in the way Dexcom does. Access to Libre data is via partnership arrangements, and those are commercial conversations, not a signup form.
There is a widely circulated unofficial approach involving the LibreLinkUp follower service. Do not build a product on it. It is not a supported interface, it can change or be shut off without notice, using it likely conflicts with Abbott's terms, and you would be putting a health product's core data path on an undocumented dependency. We have turned down this work and we'd advise anyone else to.
If you need Libre data, the practical routes are HealthKit/Health Connect (route 1), an aggregator that holds its own arrangement (route 4), or a direct commercial conversation with Abbott.
Route 4: aggregators
Terra, Spike, Vital and similar aggregators offer CGM connections alongside their wearable coverage, and they may hold vendor relationships you don't. Same trade as with any aggregator: faster start, one schema, one webhook, a recurring per-connected-user fee, less control, and they cannot give you a freshness or an access level the vendor doesn't give them.
Ask two specific questions before you sign: which CGM vendors are actually live (not "supported" on a roadmap page), and what the data delay is per vendor. The answer to the second one is frequently the same delay as route 2.
What CGM data actually looks like
Things worth knowing before you design a schema:
- Interval readings, not events. Readings arrive at a regular cadence — on the order of every one to five minutes depending on the system — which means a single user generates a lot of points. A month of one user's data is a bigger table than a year of their workouts.
- Units. mg/dL in the US, mmol/L nearly everywhere else. Store one canonical unit, format at the view layer, and never let a raw number reach the UI unformatted. Showing a US number to a UK user is not a rounding error, it's a nonsense value.
- Gaps are normal and meaningful. Sensor warm-up after application, sensor changes, out-of-range periods when the phone isn't nearby, compression lows from lying on the sensor overnight. Your charts must render gaps as gaps rather than interpolating across them, and your aggregate metrics need a data-sufficiency rule — a "time in range" computed over 40% coverage is a misleading number presented confidently.
- Sensor sessions. Sensors run for a fixed number of days, then end. Session boundaries matter for analysis and for the "your sensor expires tomorrow" experience, and they're one of the things you only get from the vendor API, not from HealthKit.
- The standard metrics. Time in range against the commonly used target band, glucose management indicator, coefficient of variation, mean glucose. If you compute these, compute them the way the clinical literature does and label the range you used, because your users will compare your number to another app's.
The regulatory line, stated plainly
CGM data is the fastest way for a wellness product to walk into medical-device territory. The boundary is roughly this: displaying and analyzing retrospective data for general wellness is one thing; alarming on values, or advising on treatment, is another. Insulin dosing guidance, hypoglycemia alerts, and anything a user could act on medically move you toward a regulated product with a clearance pathway.
This is not a disclaimer you can write your way out of. If your product design includes an alert, treat that as a regulatory decision taken deliberately, with specialist advice — start from is my wellness app a medical device. Zee Palm maps the intended-use boundary during planning and is explicit that we do not determine device status or provide regulatory sign-off.
The corollary is a genuinely good product strategy: the retrospective, correlational product is both safer and more differentiated. Nobody needs another app that shows a glucose number. What people will pay for is "this meal did this to you," "your sleep predicts your morning curve," "here's what changed since you started the medication." All of that works on delayed data, and none of it requires you to become a device.
The over-the-counter shift
The prescription-free CGMs that arrived from both major manufacturers changed who your user is. They're not managing diabetes; they're curious, they're on a GLP-1, they're optimizing. That changes the product: less clinical framing, more coaching, more correlation, more of a habit loop. It also changes churn — a curiosity-driven user wearing sensors they pay for out of pocket needs a reason to buy the next one.
If that's your audience, the adjacent piece is GLP-1 companion app development, and the build side sits under wellness app development.
What we'd build, and what it costs
A CGM integration is scoped like any other platform: $3,500 per platform, two to four weeks, covering integration design, implementation, edge-case handling, real-device QA and documentation. Details on the wearable integrations page and every price on the pricing page.
The recommendation we give most often, including when it makes our project smaller: start with route 1. Read glucose through HealthKit or Health Connect as part of an integration you probably need anyway, ship the retrospective analysis product, and find out whether users engage before you commit to a vendor relationship, a delay-tolerant architecture and a regulatory conversation. Tell us what you're building and we'll tell you which route your product actually needs.
Frequently asked questions
Is the Dexcom API free to use?
Dexcom operates a developer program with a sandbox you can build against without a commercial arrangement. Deeper access — particularly real-time data — involves a separate partner relationship. Treat commercial terms as something to establish with Dexcom directly rather than assume.
Can I build a real-time glucose alert app?
Not on the general developer API, because the data it returns is delayed by design. Real-time access requires a different relationship with the manufacturer, and alarming on glucose values also raises medical-device questions that need specialist advice before you build. Retrospective analysis is where third-party products can operate cleanly.
Does Abbott have a public API for FreeStyle Libre?
Not in the open, self-serve sense. Libre data access is a partnership conversation. The unofficial follower-service approach that circulates online is not a supported interface and is not a safe foundation for a product.
Can I just read glucose from Apple Health instead?
Often yes, and it's the cheapest and fastest route. The manufacturers' consumer apps can write glucose into the phone's health store, so a normal HealthKit or Health Connect integration reads it with no vendor partnership. You lose sensor session metadata, trend arrows and manufacturer-computed metrics — which is an acceptable trade for a first version.
How much data does a CGM user generate?
A lot more than a step counter. Readings at one-to-five-minute intervals produce hundreds of points per user per day, which makes storage strategy, downsampling for charts and query performance real design decisions rather than afterthoughts. Plan for time-series aggregation at write time rather than computing daily summaries on demand.

