The privacy terms, decoded
Term | What it means Local-first | Your files, vault and sessions stay on your machine; nothing is uploaded just to exist. Zero data retention | The gateway treats each request as ephemeral — nothing is stored after the answer. Air-gapped | Fully offline mode using a local model; no network access at all. No API keys to paste | Auth is server-side, tied to your ssv.asia account — nothing to leak in config files.
Why this matters for real work
- NDAs and client data — regulated industries (legal, healthcare, finance, defense) often cannot send documents to cloud APIs. Local-first is the answer.
- Researchers — unpublished work stays unpublished until you choose to share it.
- Air-gapped environments — the /offline workflow configures 100% offline use with a local model.
- Portability — the Expert AI Terminal Pro (separate desktop product) carries the whole installation on a USB; unplug it and nothing remains on the machine.
Honest caveat: "local-first" does not mean "never touches the internet". When you run online models, requests go to the model provider. The design minimizes exposure; it does not eliminate it for online routes. Use the air-gapped mode for truly sensitive work.
Data sovereignty in practice
- Your data is stored locally by default. ssv.asia does not upload proprietary project data to public clouds.
- Train custom models on proprietary data. Subscribers can fine-tune on their own information without sending it off-machine — full data sovereignty.
- Zero data retention. No provider API keys live on your machine.
- Air-gapped assembly. The whole system runs as a local-first workspace inside your own shell.
Who this is built for: researchers under NDA, regulated industries, and anyone who cannot afford a data leak. As ssv.asia puts it: "sovereign data hosting for rigorous domain expertise."
Good privacy habits
- Use air-gapped mode for anything under NDA.
- Check expert-ai status to see which route is active.
- Keep your Obsidian vault indexed with /lib — local search never leaves your machine.
- Prefer cited workflows (deepresearch) so you can verify what the model used.