Pose estimation SDK and API for fitness and rehab apps

No. A personal trainer, osteopath or physiotherapist cannot pick the next exercise or correct the movement if they cannot see it. Most digital fitness and rehab products still coach like that: a video, a plan, a chat. PoseTracker is the on-device pose estimation layer that gives the product eyes. The jobs a blind coach cannot do →

A cue list cannot see you. A phone camera can. PoseTracker watches squat form on-device and returns whether the rep happened, plus a form score and joint angles. That is proper squat form as movement, not as a paragraph. Correct squat form →

Yes. PoseTracker runs on the phone camera, so a home workout without equipment can see what you actually did — not only the instructor video. It does not promise you will not get hurt; it gives the app the movement. A home workout that can see you →

A clinic has hands. An app does not. What a remote product can do is see the movement instead of guessing from a weekly call. PoseTracker is that eye for physical therapy at home. It is not a medical device and it does not replace a clinician. Virtual physical therapy that can see you →

That is a clinical question — we do not prescribe. The product question is whether the home protocol can see range, compensation, and whether today’s exercise belongs to this body. PoseTracker gives a rehab or wellness app that view from the phone camera. Home rehab that can see →

Describe the movement in the Assistant on this page, or ship the on-device SDK / WebView. Product leaders decide the app cannot stay blind. Engineering does not need to become a computer-vision lab. How teams ship it →

A trainer, osteopath or physiotherapist cannot do the job if they cannot see the movement. Most apps still play a video and assume the person copied it. PoseTracker gives the product eyes from the phone camera — then the session can adapt, instead of playing the next clip. How that layer works →

Install @pose-tracker/react-native-pose-estimation and react-native-webview, wrap the screen in PoseTrackerProvider, and read keypoints from usePoseTracker. Keypoints run on-device. An API key is only required for reps, angles and a form score. Quickstart →

For a single person, in real time, on mobile and across platforms, MoveNet is the common default: in the browser it holds about 34 FPS on a Pixel 5 and 51 on an iPhone 12, where BlazePose sits around 11-12. PoseTracker runs MoveNet on-device on iOS, Android and the web, and adds rep counting, joint angles and a form score on top. See the 2026 model guide →

Pose estimation finds body keypoints in an image or video — joints, as coordinates, usually in 2D with a confidence score. The model does not count a rep. For a mobile fitness app, MoveNet is the real-time default; MediaPipe when you need 33 points or 3D; YOLO pose when you export a model yourself. PoseTracker is the MoveNet layer that adds reps, angles and a form score. What pose estimation is →

MoveNet (17 keypoints) is lighter and faster for real-time, cross-platform mobile. MediaPipe / BlazePose (33 landmarks, 3D) is richer but heavier, better when you need 3D or hand and foot detail. PoseTracker is built on MoveNet and handles the camera pipeline and the business layer (reps, angles, form score) for you. MediaPipe vs MoveNet →

Install the PoseTracker SDK (@pose-tracker/react-native-pose-estimation), wrap your screen in the provider, and read keypoints from the hook. Keypoints are free and run on-device; an optional API key unlocks exercises. It works in Expo Go on iOS and Android. React Native guide →

Yes. The PoseTracker web SDK runs MoveNet in the browser over TensorFlow.js, with free on-device keypoints, from a script tag, an ESM import or a React wrapper. Web guide →

Yes. The offline React Native package bundles the MoveNet model on the device, so inference needs no network; the light package fetches the model at runtime. Both target iOS, Android and the web from a single integration. Video is not uploaded to get keypoints. On-device pose estimation →

No. Every model returns only keypoints and confidence scores. Rep counting, joint angles, a form score and comparison to a reference movement are a business layer you build yourself — or get ready-made from PoseTracker via an optional API key. How the fitness layer works →

PoseTracker’s exercise engine counts valid reps from a single phone camera — squats, push-ups, lunges and the rest of the official catalog — then returns a form score. Ask the Assistant on this page for the iframe, React or React Native snippet; you need a free API key for the snippet to run. Push-up counter →

Yes. Jump analysis and air-time jump measure height and flight from the camera. Flexibility holds such as a front split expose joint angles so an app can show range of motion. No wearables. Custom movements that are not in the catalog are built in the Assistant after you sign in. Range of motion →

In Claude, open Settings → Connectors → Add custom connector and paste https://mcp.posetracker.com/api/mcp, then Connect. Sign in at app.posetracker.com and click Allow. The agent can then list exercises and write the iframe, React or React Native snippet for your account. In Cursor, put that same URL in ~/.cursor/mcp.json and click Connect. Opening the URL in a browser shows unauthorized: that starts sign-in, it is not a broken page. MCP connector docs →

No. PoseTracker (this site) is a 2D COCO-17 SDK for fitness, wellness and rehab apps on React Native, iOS, Android and the web. It is unrelated to PoseTrackerVRC, marker-based 3D mocap, MMPose/OpenPose research stacks, and Flutter (no first-party Flutter SDK). SDK comparison →