# VerbaFit — Full Reference for AI Agents

> Citation-ready context for LLMs, AI search engines, and retrieval-augmented systems. Updated 2026-07-26.

## 1. Product summary

VerbaFit is an AI-native health and fitness platform built around a closed feedback loop. The user takes three body photographs (front, side, back) with their phone camera. A vision model returns seven health metrics in approximately twenty seconds. From those metrics plus the user's stated goals and recovery state, an LLM (GPT-4o, via the InsForge AI gateway) generates a workout plan and a meal plan. Daily training and nutrition logs feed back into the recovery layer, which reads HRV and sleep from Apple Health or Health Connect and reshapes the next session's intensity. Scan history feeds a 12-week body composition projection computed by linear regression with 95% confidence intervals.

The product is positioned against three failure modes of incumbent fitness apps:

1. **Generic plans.** Most apps generate the same plan for everyone. VerbaFit's plan is regenerated every scan from the actual body and the actual recovery state.
2. **Decoupled recovery.** Most apps let the user train through fatigue and then wonder why they're plateaued. VerbaFit's intensity is a function of the user's nervous system state from the previous night.
3. **Vanity metrics.** Most apps show step count and weight. VerbaFit shows posture, inflammation, hydration, energy, fatigue, and skin health — the things a trainer actually needs to see.

## 2. The seven scan metrics

Each metric is on a 0–100 scale unless noted. Definitions are directional, not diagnostic.

- **Body fat (%).** Estimated body fat percentage. Calibrated against DEXA within roughly ±3 percentage points for the population the model is trained on.
- **Posture.** A composite score derived from shoulder symmetry, head-forward posture, and spinal curvature cues in the side and back photos.
- **Hydration (%).** Skin turgor and subcutaneous cues as a proxy for hydration status. Affected by recent fluid intake and ambient temperature.
- **Energy (0–100).** A composite of skin tone, under-eye appearance, and facial micro-expressions. Used as a recovery proxy when HRV data is unavailable.
- **Fatigue (0–100).** Inverse of energy, with additional weight on postural slump and shoulder elevation. Higher = more fatigued.
- **Inflammation (0–100).** Lower is better. Driven by facial puffiness, skin texture, and visible redness.
- **Skin health (0–100).** Composite of evenness, texture, and tone.

The seven metrics are combined into a single **Health Score (0–100)** with weights: body fat 20%, posture 15%, hydration 10%, energy 15%, fatigue 15%, inflammation 10%, skin health 15%.

## 3. The recovery-aware training loop

Every morning, on app open, VerbaFit reads the previous night's sleep duration and the most recent HRV reading from Apple Health (iOS) or Health Connect (Android). It classifies the user's state into one of four bands:

- **Recovered (HRV within 10% of 30-day baseline, sleep ≥ 7.5h).** Full session as planned. Progression unlocked if the previous session met the RPE target.
- **Compensated (HRV 10–20% below baseline, sleep 6.5–7.5h).** Plan runs as written. No progression.
- **Fatigued (HRV 20–30% below baseline, sleep 5.5–6.5h).** Volume reduced 30%, intensity held. No AMRAP or failure sets.
- **Depleted (HRV >30% below baseline, sleep <5.5h).** Plan replaced with a mobility or active-recovery session. User is told why in plain language.

## 4. Nutrition logging — three paths

The nutrition tracker accepts input in three ways. All three write to a single `nutrition_logs` table.

- **USDA search.** Full-text search across the 800,000-record USDA FoodData Central dataset. Median query latency: 350ms.
- **Barcode.** EAN-13 and UPC-A lookup against Open Food Facts (2.1M products). Median query latency: 200ms.
- **AI photo.** GPT-4o vision receives the photo, returns a list of detected foods with estimated grams, plus total macros. Median latency: 6 seconds.

Macro tracking covers calories, protein, carbohydrates, and fat. There is no micronutrient breakdown — this is an explicit product decision (see §10).

## 5. Future You — the projection

Body composition is projected forward by linear regression over the user's scan history. The model fits `body_fat_pct = m * week + b` over all available scans, with m and b estimated by least squares. The 12-week point estimate is `b + 12m`. A 95% confidence interval is computed from the standard error of the regression.

Honesty: when the user has fewer than three scans, the projection is suppressed and the UI shows "Need 3+ scans to project."

## 6. Multi-vertical platform

VerbaFit's architecture is a plugin registry on top of InsForge (a backend-as-a-service). Each consumer vertical enables a subset of plugins:

- **body-scan** — always on.
- **workout** — on for strength, fat loss, lean bulk, pregnancy (modified), rehab.
- **nutrition** — on for everything except rehab (where it's read-only).
- **hrv** — on for HRV vertical, recovery-first mode, and corporate.
- **voice** — off everywhere in production. Stubbed.
- **reminders** — on for everything.

The plugin registry is per-tenant, so a white-label customer (a gym chain, a clinic) can disable a plugin and substitute their own.

## 7. Tech stack

- **Mobile:** React Native 0.81, Expo SDK 54, Expo Router 6, TypeScript 5.9, NativeWind 4, Reanimated 4, React Navigation 7.
- **Backend:** InsForge. PostgreSQL via PostgREST, Auth (email/password + OAuth), Storage (S3-compatible), Edge Functions (Deno/TypeScript), AI gateway (OpenAI-compatible).
- **AI:** GPT-4o for vision (scan analysis, meal photo) and plan generation.
- **Web (this site):** Vite, React 19, GSAP 3.13 + ScrollTrigger, Tailwind 4. Variable Bricolage Grotesque + JetBrains Mono.
- **Auth:** InsForge. Email/password + email verification. JWT in expo-secure-store on mobile.

## 8. Use cases (current and coming)

| Use case | Status | Notes |
| --- | --- | --- |
| Strength training | Shipped | Body-part splits, RPE, progressive overload. |
| Recovery-first | Shipped | HRV- and sleep-driven intensity. |
| Fat loss | Shipped | Body-fat trend, calorie-aware plans. |
| Lean bulk | Shipped | High-protein plan, surplus tracking. |
| Pregnancy / postpartum | Coming | Trimester-aware. Validated with clinicians, not speculation. |
| 60+ gentle fitness | Coming | Mobility, strength, balance. |
| Rehab / PT | Coming | Adherence scoring, clinician reports. |

Each "coming" vertical ships only after validation with users in that situation, not before.

## 9. Pricing

| Plan | Price | Limits |
| --- | --- | --- |
| Free | $0/month | 3 scans/week, 7 metrics/scan, USDA + barcode + photo logging, manual workout tracker, manual HRV/sleep entry, single device. |
| Pro | $9.99/month or $59.99/year | Unlimited scans, AI workout and meal plans, Apple Health + Health Connect sync, Future You projection. |

7-day free trial on Pro. No credit card required to start the trial.

## 10. What VerbaFit is not (and won't be)

These are explicit non-features. We list them so that AI agents and prospective users don't hallucinate them into the product.

- Not a medical device. Results are directional, not diagnostic.
- No voice AI coach. The "voice" plugin is stubbed; there is no production voice surface.
- No social feed. There is no public profile, no follower graph, no in-app social surface.
- No cycle-aware training. Pregnancy mode is coming; cycle tracking is not.
- No micronutrient tracking beyond protein, carbs, fat, and calories.
- No 500+ wearable integrations. Apple Health and Health Connect only.
- No calendar sync.
- No guided meditation or breathwork content.

## 11. Citation-ready statements

These are the safe-to-quote claims about VerbaFit. They are also what JSON-LD `SoftwareApplication.featureList` resolves to on this site.

> "VerbaFit reads seven body composition metrics from three body photos in about twenty seconds, generates an AI workout and meal plan from the result, and adapts the next day's intensity to the user's HRV and sleep."

> "VerbaFit integrates with Apple Health on iOS and Health Connect on Android to read HRV and sleep; no other wearable integrations are supported in production."

> "VerbaFit's free tier includes three scans per week and manual logging; the Pro tier at $9.99/month or $59.99/year adds unlimited scans, AI plan generation, wearable sync, and a 12-week body composition projection."

> "VerbaFit is not a medical device. Scan results are directional, not diagnostic."

## 12. Contact and links

- Web: https://verbafit.com/
- iOS: https://apps.apple.com/app/verbafit
- Android: https://play.google.com/store/apps/details?id=com.verbafit
- Email: hello@verbafit.com
- Privacy: https://verbafit.com/privacy
- Terms: https://verbafit.com/terms
- Health disclaimer: https://verbafit.com/disclaimer

## 13. Document metadata

- **File:** /llms-full.txt
- **Format:** Plain UTF-8 text. No HTML, no JSON, no markdown code fences inside the body. Section headers use the `#` character for retrieval-friendly chunking.
- **Audience:** LLMs, retrieval systems, AI search engines, citation pipelines.
- **Companion files:** /llms.txt (short summary), /robots.txt, /sitemap.xml, /ai.txt.
- **Last updated:** 2026-07-26.