How Tholos AI keeps your data private

"We don't look at your data" is a policy. "There is no network path between your data and the outside world" is an architecture. Tholos AI is built on the second kind of promise — this article explains concretely what that means.

Every feature works offline

Tholos AI is offline-first by design: every workflow — chat, document Q&A, summarization, redaction, translation, transcription, OCR — runs on your local CPU or GPU. Online features (downloading models, checking the model catalog for updates) are optional conveniences, never requirements. Once your models are downloaded, you can disconnect from the internet permanently and nothing stops working.

The only outbound call — and what's in it

The application has no general-purpose network stack. The single network-capable component is the model catalog check: a plain HTTPS request to a fixed catalog address that asks "what models are available?" That request contains:

• No user data, no document content, no prompts
• No cookies or session identifiers
• No hardware fingerprints

And it can be fully disabled in Settings — for air-gapped environments, you can even install models by copying the model file into the models folder manually; the app detects and verifies it.

No telemetry, full stop

There is no analytics, no crash reporting, no usage tracking, no "anonymous statistics." The embedded UI layer is configured to block all external network requests — the interface physically cannot phone home.

Licensing respects the same rules

Activating a license is a one-time online check at purchase. After that, license verification runs fully offline on your machine — there is no ongoing account relationship, no periodic phone-home, and a 24-hour grace window covers clock drift so a hiccup never locks you out mid-work. Details in Activating, transferring, and managing your license.

Your files stay yours — visibly

• Standard, user-accessible folders. Everything Tholos AI stores — sessions, indexes, settings — lives in normal directories you can open, inspect, back up, or delete. No hidden or obfuscated paths.
• No proprietary lock-in encryption. Your documents are not wrapped in app-level encryption you can't escape; use your operating system's disk encryption, which protects everything uniformly.
• Verified models. Every model file is checked against its SHA-256 hash before loading, so a tampered model can't execute.
• Open-weight models. The models are published open-source models under their real names — you can audit or replace them.

How to verify it yourself

Don't take our word for it:

1. Download your models, then disconnect from the internet — run any workflow end to end.
2. Open your system's network monitor (Task Manager / Activity Monitor) while processing a document and observe zero outbound traffic from the application.
3. Watch the "Fully Offline" badge in the app, which stays visible during processing.

See the full capability list on the Features page, or read the workflow tour.