ATMO SDK: add stress regulation
to your product

We built the closed-loop engine. You ship it to your users.

Modular SDK. Measure heart rate and stress from camera. Guide the user through breathing, acupressure, and haptic cues. Measure the effect. All on-device — no cloud, no data leaves the phone.

iOS Android On-device AI Zero cloud 30–60 FPS PPG

Modules

Pick what you need

PPG Engine
Camera-based heart rate and HRV measurement. Fitzpatrick I–VI skin calibration with adaptive thresholds. Honest null when signal insufficient. C++ signal core — 94 KB.
BPM accuracy: <3 BPM MAE vs ECG
RMSSD correlation: >0.85
Frame latency: P95 ≤30ms
Reference: Polar H10
State Engine
Real-time classification: calm, fatigue, stress, recovery. Rule-based, no ML dependency. Circadian-aware. Optional SLM generates human-readable explanations in 4 languages — not required for core logic.
Classification: deterministic rules, <5ms
Text generation (optional): Qwen 2.5 0.5B (300MB)
Inference: <3 sec (text only)
Fallback: template strings, instant
Prediction: 2–4h ahead (logistic regression, after 14 days)
Intervention Engine
28 breathing protocols, 15+ acupressure points with illustrations, haptic guidance patterns. Adaptive — learns which protocol works for each user.
Selection: Thompson Sampling (MAB)
Context: time, circadian, season, lunar, TCM
Session: 3 min average
Evidence: PubMed 2020-2025
Learning Engine
On-device personalization. System improves weekly. Tracks what works, when, for whom. Stress prediction trains on personal data.
Training: weekly, during charging
Storage: 90 days, AES-256
Rollback: auto if model degrades
Inputs: HRV + calendar + tags + rhythm

Architecture

Closed loop, not content library

ATMO closed-loop architecture: Sense → Interpret → Intervene → Validate → Learn → repeat
// Integration example (conceptual) final atmo = AtmoSDK.init(config: AtmoConfig( modules: [.ppg, .stateEngine, .intervention, .learning], locale: "en", sensorSource: .camera, // or .healthKit, .polar, .custom )); // Start a session final session = await atmo.startSession(); // Get result when done session.onResult((result) { result.state; // .stress | .calm | .fatigue | .recovery result.bpm; // 78 result.hrv; // 42.3 (RMSSD ms) — null if insufficient signal result.intervention; // recommended protocol result.delta; // pre/post HRV change });

Use Cases

Who integrates this

Corporate Wellness
3-minute stress check between meetings. No employee data on your servers.
Fitness / Recovery Apps
Post-workout HRV + guided breathing. Prove recovery, not just track it.
Telehealth
Remote vital signs + intervention protocols between appointments.
HR-Tech
Burnout prevention layer. Detects patterns, suggests action. Zero surveillance.
Wearable OEM
Your hardware + our intervention intelligence. Haptic-only, no screen needed.
Insurance / Preventive Health
Engagement tool that reduces claims. On-device = no data liability for you.

Why This Architecture

On-device means zero liability for you

Your user's data

  • Never hits your servers
  • Never hits our servers
  • Never hits any server
  • You are not a data processor

Your costs

  • No cloud inference: $0/user
  • No storage per user: $0/month
  • No API rate limits: unlimited
  • Scales to 1M users: same cost

Your compliance

  • Health data stays on-device
  • No data processor liability for you
  • No server to breach
  • No user data to subpoena

Specs

For your engineering team

PPG Processing

  • Frame P95: ≤30ms
  • Frame P99: ≤40ms
  • Capture: 30–60 FPS adaptive
  • Core: C++ arm64, 94 KB
  • BPM MAE: <3 BPM

AI Model

  • Qwen 2.5 0.5B: 4-bit GGUF
  • On-disk: ~300 MB
  • RAM: ~400-500 MB
  • Inference: <3 sec
  • Timeout/fallback: 5 sec

Platform

  • iOS: 15.0+ (Core ML)
  • Android: API 24+ (TFLite)
  • Flutter: 3.35+
  • Native: Swift / Kotlin
  • Encryption: AES-256 at rest

Ready to integrate?

Early access available. Tell us your use case.

Request SDK Access Ask a Question