Security
Boring, in the best way.
No clever marketing language, no asterisks. Here is exactly what we do with your data and how we protect it.
You connect it, one piece at a time
Nothing is assumed. You connect each integration — Garmin, WHOOP, your calendars, Telegram — explicitly, one by one, and every one is optional and disconnectable under Settings → Integrations. While kynikOS is in beta it auto-approves the actions you ask for; you tune that behaviour under Settings → Behavior.
Keys the database never holds
Everything travels over TLS 1.2+. Your conversations are encrypted app-side with AES-256-GCM, and the keys live outside the database provider — in the app host's environment. So neither a leaked database credential nor the database operator can read them. Wearable OAuth tokens get their own dedicated encryption, and the underlying disks are encrypted too.
A plain-language history of what we did
Every action kynikOS takes on your behalf — bookings placed, calls made, events created — is written to a history you can open, read, and search in the dashboard. No obfuscation, no black box.
Hosted in Switzerland, watched on our own infra
Your database lives in Switzerland (Zurich) — which the EU recognises as an adequate destination for GDPR data. App services run on Railway. Our tracing runs on our own infrastructure, so no third-party analytics vendor sees your prompts, and traces carry metadata only — not message content. Error tracking is configured to exclude personal content. GDPR is the floor, not the ceiling.
Health data
Never sent to outside AI.
Your raw wearable metrics — sleep, HRV, readiness, activity, body composition — never reach any external AI. Four redundant layers enforce it: a restricted database role that can only read model-safe context, REVOKE/GRANT and row-level filtering that strips anything not marked safe, a CI check that pins every model call to a single reviewed code path, and a compile-time egress contract on every agent tool. Only derived categorical bands — “sleep: poor”, “recovery: low” — are ever allowed through. Full details are in our privacy policy.
What the model reads
The honest part.
To act on your behalf, the agent has to read what you tell it: your free-text messages, your calendar event titles, and the personal context you choose to declare — including life-phase facts. That text is processed by Anthropic in the US. It is API traffic: Anthropic does not train on it. We keep what reaches the model to what the task needs, and the biometric gate above still applies to everything from your wearables.
Export and deletion
Ask and we handle it.
There is no self-serve export or delete button yet, and we will not pretend otherwise. Email [email protected] — for a copy of your data, deletion, or any general or legal question — and we handle it promptly, by hand. Deleted data also ages out of our backups as they roll off, within about thirty days.
Responsible disclosure
Found something?
If you discover a security vulnerability in kynikOS, please report it to [email protected]. We acknowledge all reports within 48 hours and aim to assess and respond to the substance within 7 days. We ask that you give us reasonable time to address confirmed issues before any public disclosure.
We do not offer a formal bug bounty programme at this time.