Open to inspection.

Access comes before retrieval.

Matter, document and knowledge access follow each user’s tenant, workspace and permissions. Search applies that scope before it retrieves private records, rather than fetching everything and hiding it afterwards. The platform operator is not automatically entitled to read case records.

Different people, different acts.

Where AI processing happens.

Today, model calls run outside India. Most go through a zero-data-retention route with data collection denied. Gemini and GPT calls go directly to Google and OpenAI under their API terms; Google keeps that data for a limited period for abuse monitoring. The route that served each call is recorded on the run, and embeddings use a separate self-hosted model. Hosting location and AI processing location are separate questions; confirm both for your deployment.

Decisions leave a trail.

Draft versions and approvals carry a content hash, so an approval applies to one exact text. Model runs keep their review record. Administrative changes, exports and privileged access are audited.

Retention is set per tenant.

Tenant policies, document classes and legal holds decide what is kept or deleted. Zero-retention model routing doesn’t mean your own case record or audit trail disappears.

Authentication follows your workspace.

Organisation single sign-on uses OpenID Connect and a verified email domain, and membership must already exist. A federated sign-in never creates a workspace or grants access to a matter. SAML, SCIM and government identity providers should not be assumed available yet.