Why Mandrel?
As AI agents move from “chatbots” to “autonomous actors,” the need for rigorous, machine-readable governance becomes critical. Traditional security models are ill-suited for the fluid, non-linear execution patterns of agentic systems.
Mandrel bridges this gap by providing Zero-Trust Enforcement for AI agents, ensuring they operate strictly within their defined operational boundaries and risk envelopes.
Design Philosophy
1. Declarative Governance
We believe governance should be a contract, not an afterthought. Every Mandrel agent is bound to a Mandrel Spec that defines its identity, capabilities, and constraints.
2. Zero-Trust by Design
Every “turn” in an agentic workflow is intercepted and verified. No agent is trusted implicitly; every action must be justified against its governing matrix.
3. High-Performance Enforcement
Security shouldn’t be a bottleneck. Our enforcement proxy, the Mandrel Collet, is engineered for sub-millisecond overhead to maintain real-time agent responsiveness.
Component Ecosystem
Mandrel Spec
The declarative contract for autonomous agent governance. Language-neutral schemas for operational boundaries.
Mandrel Collet
High-performance enforcement proxy (sidecar) that “clamps” agents to their specifications.
Mandrel CLI
Governance tooling for validating specs and generating deployment artifacts.
Mandrel Gantry
Kubernetes Operator for agent lifecycle management and SPIFFE identity provisioning. Coming soon.
Metrology Lab
Attestation registry and semantic discovery gateway for the governed agent mesh. Coming soon.