Security & architecture
Mantid is an edge-first AI system. Inference runs inside your network on a Mantid box. Raw video doesn't leave your floor by default — only metadata and alerts do. Built for safety, ops, and compliance teams in regulated industries.
How it's built
Everything below is reviewable by your security team. Detailed documentation, SOC 2 reports, and architecture diagrams are available under NDA.
01 · ARCHITECTURE
Inference runs on a Mantid box deployed inside your network. Raw video never leaves the site unless you choose to send a clip out — for incident review, audit, or shared training data. Metadata, alerts, and analytics are what go to the cloud, not your floor footage.
02 · ENCRYPTION
All stored data — model weights, configuration, event logs — is encrypted with AES-256. All network traffic between your site and our control plane is over TLS 1.3 with modern cipher suites. Keys are managed in a hardware-backed key store and rotated on schedule.
03 · ACCESS
Customer admins control who on their team can view footage, configure detectors, and download evidence. Every privileged action — log-in, export, configuration change — is recorded with user, timestamp, and IP. Logs are tamper-evident and retained for the life of the contract.
04 · COMPLIANCE
Mantid inherits Dragonfruit AI's SOC 2 control framework. We're built to integrate with customer-side SSO (SAML, OIDC), data-retention policies, and audit requirements for retail, logistics, manufacturing, and food-and-grocery operators.
05 · OPERATIONS
Our security team monitors infrastructure 24/7. Dependencies and OS images are patched on a defined cadence; critical CVEs are addressed within a published SLA. Incidents are triaged, communicated to affected customers, and post-mortemed.
06 · DATA LIFECYCLE
Customer data is segregated by tenant. On contract termination, we provide a documented data-export and deletion path. Retained training data is opt-in and anonymized — no customer footage is used to train shared models without explicit written consent.
Documentation
Talk to security
Email us with your vendor security questionnaire, architecture review request, or pen-test findings. We respond within two business days.
contact@mantid.ai