Tenant-bound intelligence: AI that never leaves your boundary
Search, OCR and embeddings run inside the workspace instance, not in our control plane. Here is why that architectural line matters more than any privacy promise.
By The SisiDesk Team
Most products that add intelligence to your files do it the easy way: ship the content to a central service, index it there, and promise not to look. We think the promise is the wrong artefact. The architecture should make the promise unnecessary.
Where the work happens
In SisiDesk, optical character recognition, embedding generation and the vector index all run inside your workspace instance, within the tenant boundary. The control plane — the part that provisions instances, meters usage and issues bills — has no read path into that data. It cannot search your documents because it cannot reach them.
- Document content, embeddings and indexes stay on tenant-scoped storage.
- The control plane sees session metadata: start, stop, region, minutes, cost.
- No cross-tenant model training. Ever, on any tier.
The trade we accept
This design costs us things. Features are slower to ship because they cannot be built once centrally. Some cross-workspace conveniences are simply unavailable. We think that is the correct price for a workspace that holds the most sensitive thing a professional owns: everything they are working on.
If a vendor can read your files, the only thing standing between you and exposure is their policy. We would rather remove the capability than write the policy.