What our AI features do
- Semantic search: natural-language retrieval over your own tenant's files — only files the caller is allowed to read are ever searchable.
- Grounded answers: an agent answers questions using only your own indexed files, and every answer carries the citations it was grounded in.
- Content indexing: after a file passes the malware quarantine gate, its text is extracted, chunked and embedded for retrieval. Only clean, production content is indexed.
Providers and models
Today, every AI provider in the serving path is a deterministic, in-process algorithm operated by us — no third-party or hosted general-purpose model processes your content. When hosted models are introduced, they will be listed here and in the per-tenant transparency report before they are enabled, with residency-safe routing pinned to your tenant's region and PII redaction applied before any egress.
Tenant administrators can query the exact providers and models in use at any time via the API (GET /v1/ai/transparency).
Your controls and our commitments
- Per-tenant kill switch: an administrator can disable all AI features for the tenant in one action; indexing, search and answers stop immediately.
- Every AI action is written to the tenant audit trail — recorded by shape (sizes, counts, latency), never by query or answer content.
- Your content is never used to train or improve any model.
- All AI processing is tenant-isolated and never crosses tenant boundaries.
EU AI Act posture
Our AI features — semantic retrieval and grounded question answering over a customer's own documents — do not fall within the high-risk use cases enumerated in Annex III of the EU AI Act.
In line with the Article 50 transparency obligations, AI-generated answers are explicitly labeled and always accompanied by their source citations. We will re-assess this posture and update this note before any hosted general-purpose model is enabled in the serving path.