# Somvia

> Somvia is an iPhone app that analyzes sleep, recovery and training from Apple Health. Scores are calculated on the iPhone. Users can optionally share selected health data, scores, workout details and saved memories with compatible AI Agents through remote MCP.

Which Somvia this is: the iPhone app at https://somvia.app, listed on the App Store as "Somvia: Training Load & MCP" (app ID 6786936154) and made by an independent developer. It analyzes training load, recovery and sleep from Apple Health and runs a remote MCP server for AI Agents. It is not a sleep-sound, meditation or relaxation app, and it has no Android or Google Play version.

Key facts:

- App Store: https://apps.apple.com/app/id6786936154 (iOS; free to install, Somvia Pro subscription or lifetime purchase unlocks hosted sync and agent access)
- MCP endpoint: https://somvia.app/mcp — remote MCP over HTTPS. ChatGPT, Claude, Codex, and other OAuth-capable clients sign in with Apple. OpenClaw, Hermes, Grok, Muse, and other agents that connect with a key can use a private `som_` bearer access token issued in the app.
- Local analysis: recovery and sleep scores, HRV status, training load and status, workout analysis, trends, threshold estimates and race predictions. Available results depend on the records in Apple Health; each analysis has its own data requirements. Existing records count toward the required history.
- Shared context: users and Agents can save goals, conditions and preferences as memories, and attach notes to workouts. Saved context is available to connected Agents and can be reviewed, edited or deleted in the app.
- Available data: sleep and sleep stages, heart rate, resting heart rate, HRV, workouts, steps, active energy, vitals, and other HealthKit metrics the user explicitly enables. Workout detail can include splits and heart-rate, speed, route, power, cadence, or stroke-rate series when Apple Health records them.
- History: the first sync reads back 7, 30, 90, or 180 days of daily metrics (the user chooses) and 180 days of workout sessions and available detail series. New data keeps syncing after that and is not deleted by age.
- Privacy: health data upload is off by default; local analysis reads data authorized in iOS; sync selections control uploaded categories separately; data encrypted in transit; access tokens stored server-side only as SHA-256 hashes; no ads; no data resale; account and synced data deletable in-app.
- Model: Somvia runs no language model of its own. The user's own agent answers with the model the user selects, so users can use the strongest model available to them, and the model only changes when they change it. Scores are calculated on the iPhone with published formulas, and the server does not call a language model. There is no AI allowance or credit system.
- Identity: Sign in with Apple. Reinstalls and new devices recover the same account, data, and subscription.

## Pages

- [Home](https://somvia.app/): product overview, how it works, supported clients, FAQ
- [Apple Health MCP](https://somvia.app/apple-health-mcp): remote MCP architecture, data depth, export/local-server comparison, a product-by-product comparison of Apple Health MCP servers (Somvia, HealthAPI, AthleteData, Attune, Wellness Project, HealthKite, MetricBridge, vpetersson/apple-health-mcp-server, Open Wearables) checked on 2026-10-19, and client guide index
- [Why I built Somvia](https://somvia.app/why-i-built-somvia): how the developer uses an AI Agent with Somvia to plan running alongside CrossFit and HYROX; workout notes, memory, and how Claude, ChatGPT, OpenClaw, Hermes and other MCP clients connect
- [Hermes Agent + Apple Health](https://somvia.app/hermes-apple-health): bearer-token setup in `~/.hermes/config.yaml`, verification, and token safety
- [OpenClaw + Apple Health](https://somvia.app/openclaw-apple-health): Streamable HTTP setup and comparison with OpenClaw's native current-day HealthKit summary
- [Apple Watch workout analysis](https://somvia.app/apple-watch-workout-analysis): session, split, and time-series data for training questions
- [Claude + Apple Health](https://somvia.app/connect-apple-health-to-claude): when to use the Claude iPhone app's built-in Health access (beta, Pro and Max, US) and when to use Somvia (Claude Desktop, web, Claude Code, other countries, workout detail); remote custom connector and Sign in with Apple flow
- [ChatGPT + Apple Health](https://somvia.app/connect-apple-health-to-chatgpt): native Health versus Somvia's current custom-MCP path and availability limits
- [How the scores work](https://somvia.app/training-science): the models behind every score, what each needs, and the limits
- [Recovery score](https://somvia.app/apple-watch-recovery-score): what the morning score reads, why it needs your own baseline, and what Apple Health must contain
- [Training load](https://somvia.app/apple-watch-training-load): load per session, long-term load against this week, and the eight training states; long-term load is CTL (42-day) and the seven-day average is ATL, with TSB available to agents over MCP
- [VO2 max on Apple Watch](https://somvia.app/vo2-max-apple-watch): Apple's Cardio Fitness against an estimate from your own sessions, and how to read the gap
- [Threshold pace and heart rate](https://somvia.app/lactate-threshold-apple-watch): field estimation from hard efforts already in Apple Health, and critical power for rides
- [Race time predictions](https://somvia.app/race-time-predictor-apple-health): 5 km to marathon solved from a qualifying effort, and what the marathon assumes
- [Hyrox training](https://somvia.app/hyrox-training-apple-watch): recovery each morning, one load across runs and stations, and whether the running is getting faster; no plan, no station analysis, no Hyrox time prediction
- [Best Apple Watch recovery apps](https://somvia.app/best-apple-watch-recovery-apps): Athlytic, Bevel, Training Today, Gentler Streak and Somvia compared on what each recovery score reads, training load, threshold and race predictions, AI and agent access, and US prices, checked on 2026-10-19; how to choose, what Somvia does not do, and why Garmin Training Readiness needs a Garmin watch
- [Hybrid athlete](https://somvia.app/hybrid-athlete-training-load): running and lifting added into one training load, the two averages, and the morning recovery that includes the lifting
- [Changelog](https://somvia.app/changelog): release history covering the interface, sleep, recovery, training analysis, notifications and Agent connections
- [Privacy Policy](https://somvia.app/privacy): what is processed, how it is used, retention, deletion
- [Support](https://somvia.app/support): common checks and contact (support@somvia.app)
- [Terms](https://somvia.app/terms): Apple standard EULA, subscription terms

## How a user connects an agent

1. Install Somvia on iPhone and authorize the Apple Health data it may read. Local scores and analysis do not require health data upload.
2. Open Connection from Today or Settings, enable syncing and choose which categories to upload. Selected data becomes available through the authenticated endpoint at https://somvia.app/mcp.
3. In ChatGPT, Claude, Codex, or another OAuth-capable MCP client, add Somvia as a remote connector and sign in with Apple — or use the app-issued access token in OpenClaw, Hermes, Grok, Muse, or another agent that connects with a key.
4. Ask a health or training question. The agent can start with daily summaries, then fetch workout splits or time series only when the question needs them.

Somvia is not a medical device and does not provide medical advice, diagnosis, or treatment. Users should consult a qualified healthcare professional before making medical decisions.