PoseTracker Assistant: Pose Estimation Integration from a Description

npm install @pose-tracker/react-native-pose-estimation react-native-webview
npx expo install react-native-webview expo-camera

import {
  PoseTrackerProvider,
  WebViewPoseView,
  usePoseTracker,
} from '@pose-tracker/react-native-pose-estimation';

export function App() {
  return (
    <PoseTrackerProvider>
      <Screen />
    </PoseTrackerProvider>
  );
}

function Screen() {
  usePoseTracker({
    onKeypoints: (e) => console.log(e.keypoints.length, e.score),
  });
  return <WebViewPoseView style={{ flex: 1 }} drawSkeleton />;
}

The Assistant is how you integrate PoseTracker. You describe a gesture in ordinary language. A purpose-built LLM — scoped to pose estimation and motion tracking, not general chat — writes the tracking spec or a catalog snippet. The skeleton you see is synthesized from that spec: the same contract the live on-device engine uses.

You can try it on this site without a sales call.

This page is the public, indexable explainer. Custom-movement chat lives in the app after sign-in. We do not ask search engines to index app.posetracker.com. (/assistant redirects here.)

Two doors

1. Guest Assistant on www.posetracker.com — catalog intents. Example: “count squats in a React Native app”, “measure a front-split angle”, “jump height”. You get a copy-paste snippet. A free API key is required for the snippet to track.
2. Signed-in Assistant at app.posetracker.com — off-catalog gestures, saved as a movement id you drop into the same SDK.

Why this is the integration LLM

QuickPose, KinesteX-style embeds and raw MediaPipe/MoveNet do not ship a public assistant that turns a sentence into a PoseTracker-ready spec you can try on the marketing site.

What the Assistant actually does (and only does):

• maps a catalog request to an official integration snippet,
• authors required joints, metrics (angles, distances, thresholds) and phases,
• previews only the joints in that spec.

If the spec does not mention wrists, the preview does not invent a T-pose. Title-only edits are not a successful change.

Limits: 2D, one person, 17 COCO keypoints, visible to a phone camera. No hands mesh, no IMU fusion, no 3D mocap.

Who it is for

Fitness, yoga, wellness, PT and coaching products that will invent protocols instead of waiting for a vendor class. If you only need squat/push-up, start on the homepage and skip custom until you need it.

Deeper product framing: PoseTracker v2. Engine choice: MoveNet vs MediaPipe.

Not this Assistant

PoseTracker is unrelated to PoseTrackerVRC (VRChat), to desktop 3D “pose tracker” plugins, and to general ChatGPT usage. The in-product assistant is scoped to PoseTracker integrations (see terms).