Scoped identities
Every deployed worker receives a tenant-specific service identity, named tools and explicit limits.
HomeTrust by operating design
Authority comes from policy—not model confidence.
LatentCrew is being designed as a bounded managed service. This page states the target controls and current product commitments; it is not a certification claim.
Human authority
Consequential proposals bind exact parameters to an authenticated approver and expiry.
Visible failure
Uncertain work is reported as unconfirmed or escalated—not silently marked complete.
Tenant boundaries
Data, retrieval, memory, caches, credentials and support access remain tenant scoped.
Erasable content
Raw customer content stays out of immutable evidence and follows its approved retention policy.
Release evidence
Roles, workflows, prompts, routes and connectors require evaluation, canary and rollback records.
Truthful data-path claims
“Hosted in Mumbai” is not the same as India-only processing.
Every production route must disclose application storage, model inference, support access, provider retention and fallback geography separately.
01Application storageRegion and recovery path
02Model inferenceEndpoint and processing geography
03Provider treatmentRetention, abuse review and training terms
04FallbackOnly a separately approved equivalent route
Current status
Pre-launch trust programme
Security testing, counsel review, provider contracting and operational evidence are release gates—not claims already completed.
- PRD and threat model established
- Independent penetration test before general availability
- DPA, subprocessors and retention approved before paid beta
- Timed restore and deletion exercises before production promise