Definition
What is Sanctum Runtime?
Sanctum Runtime sits between models and execution. Every high-stakes action is evaluated against policy before it runs — locally, offline-capable, with a full audit trail.
Trusted runtime, not just a guardrail library
A trusted runtime intercepts actions at execution time: send email, move a robot, unlock a door, call an API. The runtime returns approve, verify (human-in-the-loop), or block — then records evidence for operators and compliance systems.
Runtime orchestration for embodied AI
Embodied AI and fleet deployments need more than prompt safety. Sanctum connects runtimes to a control plane: policies sync across deploys, events stream to your dashboard, and operators resolve verifications from a single console.
Agent governance
- Policy engine: approve, verify, or block per action type
- Offline heuristics and local LLM risk scoring (open core)
- Audit log, webhooks, and marketplace adapters (ROS2, agents, smart home)
- Hosted console for fleet map, billing, and enterprise SSO
Get started
Full reference: documentation · llms.txt · architecture.md
