Welcome to TeeChat
We’re building TeeChat for people who want a capable AI assistant without giving up ownership of their conversations.
Desktop (available now)
Native desktop apps for macOS Apple Silicon, Windows, and Linux are available now — chats and files stay in local folders you control, with optional local models and the full experience. Download from www.teechat.ai/download. An Intel Mac build is still rolling out. Mobile apps for iOS and Android are still in development. Your existing account carries over when you install desktop.
Data sovereignty on teechat.ai (desktop first)
Data sovereignty shows up first on desktop: conversations live in folders on your machine — your folder, you rule. We do not keep a copy of your chat threads on our servers. See the homepage FAQ.
When you use hosted AI for replies, your message leaves the device already scrambled — our routing server forwards sealed envelopes and should not see what you actually typed. Something must unscramble it to run the model: only our open-source inference engine, inside a secure hardware zone, does. Open Settings in the desktop client and verify the engine before you send — so you know which measured environment will see your prompt for that reply.
Web (hosted, paid)
A Standard or Pro seat can also use hosted chat in the browser at chat.teechat.ai — locked envelopes and engine verification in Settings. Threads stay in that browser, auto-delete after about 7 days, and do not sync across devices. Web is not on Free. Folder sovereignty is the desktop app.
Open source you can audit
The TeeChat application is proprietary. We publish the encryption stack as open source so experts can verify the hosted path — not just trust marketing:
- OPE (Open Privacy Envelope) — how we lock and sign messages
- Inference engine — verification, key rotation, and server-side AI handling
Verified AI replies (when you need proof)
Our production path uses verified inference engines: your app locks the message with OPE; the routing server forwards sealed envelopes; only the engine inside a locked-down secure server unscrambles it for the model. This is evidence you can check, layered on top of data sovereignty.
We’ll share measurement and policy details as production verification goes live.
Follow along
- Product updates on this blog
- Open protocol code: OPE and Inference engine
Revision log
- Web is a paid hosted client, not a Free preview.
- Point open-source engine links to teechat-inference-engine-rust (Rust production train).
- Desktop now on macOS Apple Silicon, Windows, and Linux; lead with desktop, de-emphasize web preview.
- Glossary: keep zh '对话' for chat thread (not 线程).
- Clarify desktop availability — macOS Apple Silicon only for now; Windows, Linux, and Intel Mac rolling out next.