Every iPhone ships with a physics lab. We turned it on.
ABPhysicsAI turns iPhone and iPad into scientific instruments — real-time sensor capture, live signal analysis, AR visualization, and a physics tutor powered by on-device AI. No cloud. No accounts. No third-party code.
Four instruments, one bench.
A complete laboratory workflow — from raw measurement to understanding — in a single app.
Measure the world around you
Motion, audio, magnetism, light, and location — captured in real time with sub-50 ms latency and a live FFT running on every window.
See the physics in your data
Raw signals become trajectories, vector fields, spectrograms, and harmonic-motion models — rendered at 60 FPS in Swift Charts and RealityKit.
Ask why, not just what
An on-device physics assistant that explains your measurements, teaches the concepts behind them, and answers questions grounded in your own data.
Run real experiments, guided
The AI designs structured experiments — setup, procedure, formulas, safety notes — validated by the app and run step-by-step with the sensors you have.
From sensor to insight, one disciplined pipeline.
Every measurement travels the same path — the order matters, and it's the same every time.
Acquire
Raw samples from CoreMotion, AVFoundation, and CoreLocation at full rate.
Normalize
Every value converted to SI units before anything else touches it.
Filter
Noise suppressed with Accelerate-backed DSP, under 10 ms per FFT window.
Extract
Peaks, means, dominant frequencies, and axes distilled into features.
Store
Sessions persisted locally with SwiftData — replayable, exportable as CSV or JSON.
Summarize
A compact feature summary — never raw arrays — is handed to the AI.
The AI never sees raw sensor streams. It reasons over small, deterministic feature summaries — which keeps analysis fast, private, and reproducible.
An AI that shows its work.
Science depends on knowing where a number came from. Every value in ABPhysicsAI carries a provenance badge — and the AI is bound by the same rules.
Read directly from hardware sensors. What the instrument actually saw, with its sampling rate on record.
Derived by formula from measured values. The formula is shown, so you can check the math yourself.
Inferred by the AI, always with a stated confidence. Never dressed up as a measurement.
- The AI never fabricates a measurement or claims certainty it doesn't have.
- Three safety layers — prompt guardrails, model filters, and output validation — check every response.
- Experiment plans are validated before they run, and unsafe procedures are refused outright.
Your data never leaves the device. Neither does the AI.
Inference runs on Apple Foundation Models, entirely on-device. There is no server, no account, no analytics pipeline, and no third-party framework anywhere in the app — privacy enforced by architecture, not by policy.
Classrooms and field work included: everything works offline, on an airplane, or in a basement lab with no signal.
| Sensor latency | < 50 ms |
| FFT per window | < 10 ms |
| Graph rendering | 60 FPS |
| AI inference | on-device |
| External AI services | 0 |
| Third-party frameworks | 0 |
| Platforms | iOS 17+ / iPadOS 17+ |
The lab opens on TestFlight.
ABPhysicsAI is entering beta for iPhone and iPad. Be among the first to run real experiments with the hardware already in your pocket.
Request beta access