Get in touch
Efferens SDK In development

Designed to add exercise detection in a few lines of code.

We're building native SDKs for watchOS and Wear OS, with a React Native bridge on top. The model is designed to run on the watch, work offline, and hand your app finished sets as typed events.

$npx expo install @efferens/react-native
Added @efferens/react-native
Config plugin: watch target, HealthKit and Health Services permissions
On-device model bundle added

Illustrative package names. Nothing is published yet; final names will be set before release.

watchOS · planned

Swift · Swift Package Manager
  • SwiftUI views and async/await APIs
  • Core ML model, designed to run on the watch
  • HKWorkoutSession for background sessions

Wear OS · planned

Kotlin · Gradle
  • Jetpack Compose components and Flow APIs
  • LiteRT (TensorFlow Lite) model, designed to run on the watch
  • Health Services exercise sessions

React Native · planned

TypeScript · npm
  • Typed hooks: useSession, useSets
  • Designed to be New Architecture (TurboModules) ready
  • Expo config plugin, designed to need no manual native setup
  • Phone and watch designed to stay in sync for you
Quickstart

The planned path from install to your first detected set.

# React Native (Expo or bare)npx expo install @efferens/react-nativenpx expo prebuild# Bare React Nativenpm i @efferens/react-nativecd ios && pod install

Illustrative API, not the final surface.

Architecture

The model is designed to run on the wrist.

Detection is designed to happen on the watch, so it can work without a signal. Your app will receive structured sets, not sensor streams. With user consent, sensor recordings go to Efferens to improve the models.

On the watch
SensorsAccelerometer · gyroscope
Efferens modelCore ML · LiteRTBack squat · rep 12
Set eventsexercise · reps · timing
Designed to work fully offline
Your app
Phone appWatchConnectivity · Data Layer
Back squat · 12Deadlift · 12
Your backendYour database, your rules
Efferens
ConsoleRollout, catalogue, quality metrics
Model trainingConsented sensor recordings
Under the hood

Designed for real gyms and real devices.

On-device and offline

Inference is designed to run on the watch with no network calls during a workout. Sets will queue locally and sync when the phone is back in range.

Phone in range3 sets synced
Back squat · 12Synced
Bench press · 10Synced
Deadlift · 8Synced

Battery-conscious design

The design goal: adaptive sampling and the platform's own workout session APIs to keep the watch efficient through long sessions. Battery impact hasn't been measured yet.

SamplinghighStateIn set
Now

Typed events

Every set is designed to arrive with the same schema on every platform: TypeScript types, Swift Codable and Kotlin data classes.

type DetectedSet = {
  exercise: "back_squat" | …
  reps: number
  confidence: number
  startedAt: Date
  durationSec: number
}

Confirm-UI components

Planned drop-in confirm views for low-confidence sets on watchOS, Wear OS and React Native: "Was that a row?" Native on each platform, themeable, or build your own.

Not sure about that oneWas it a row?
YesDeadlift

Remote settings from Console

Planned: exercise catalogue, confidence thresholds and rollout flags update over the air, with no app release needed.

cataloguev14
squat.threshold0.80
rollout25%
Synced 2 min ago
Developer tools · planned

Test without lifting a thing.

You'll record a session once on a real watch, then replay it in the simulator, in unit tests and in CI, so your detection code gets the same events every time.

  • Session recorder and simulator replay
  • Fixture library of recorded workouts for tests
  • On-watch debug overlay with live confidence
  • Planned: sample apps for each platform
  • Docs your AI coding assistant can read
Replay · squat-session-03.efr00:09
123456789101112
DetectedBack squat12 reps
✓ example: detects 12 reps of back_squat (1 passed)
FAQ

Questions engineers ask first.

Something else? We'd rather answer it on a call with your team.

Will it work with Expo?

That's the plan. The config plugin is designed to add the watch target, entitlements and permissions during prebuild, so there's no manual native setup.

Will the React Native New Architecture be supported?

That's the plan. The bridge is designed as TurboModules and to also work with the legacy architecture during migration.

Will detection need a network connection?

It's designed not to. The model is designed to run on the watch, with sets queued locally and synced to the phone when it's in range.

Does sensor data leave the device?

Detection is designed not to need it: the model is designed to run on the watch and your app will receive set events. Separately, sensor recordings, with the user's age, height and weight, will be uploaded to Efferens only with the user's consent and used to improve the models. How consent is collected is still being defined.

Which watches will be supported?

We're targeting Apple Watch on watchOS and Wear OS watches with accelerometer and gyroscope access. We plan to publish a tested-device list once there's a release.

Can we add our own exercises?

We plan to offer custom models. Your catalogue would then ship through the Console like any other update.

How will we test without a gym?

The plan is to use recorded sessions: replay them in the simulator, in unit tests or in CI and assert on the events you get back.

SDK

Build exercise detection into your app.

We're looking for early partner workout-tracking apps. Tell us about your stack and we'll share where we are.