AccuKnox, the Menlo Park cloud-security startup behind the open-source KubeArmor runtime-security project, announced AgentZ on August 27 — a platform aimed at closing the gap between enterprises experimenting with AI agents and actually running them in production. The company, which has built its business on zero-trust Kubernetes and cloud-native application protection, is positioning AgentZ as a governance-first alternative to teams stitching together separate agent frameworks, sandboxes, and permission systems by hand.

What AgentZ bundles together

AgentZ combines agent building, an execution environment, tool access, workflow orchestration, and permissions into one stack, organized around a model of Organizations, Workspaces, Agents, Workflows, and Sandboxes, with role-based access sitting across that structure for centralized administration. According to the announcement, the platform is model-agnostic, with support for OpenAI, Anthropic's Claude, xAI's Grok, and other models, and it can run agents in sandboxed, isolated execution environments rather than giving them direct, unconstrained access to production systems.

On deployment, AccuKnox is targeting the parts of the market most sensitive about where agent workloads run: AgentZ supports SaaS, on-premises, and air-gapped installations, plus bring-your-own-LLM configurations, so regulated organizations can keep model calls and data inside their own infrastructure boundary if required.

Governance as the pitch, not an add-on

AccuKnox is explicitly framing AgentZ as "an AI platform with security built in" rather than a security product retrofitted with AI features. The company says runtime credential injection, tool-level permissions, and audit trails are core to the design — capabilities meant to give security and compliance teams the same kind of controls and visibility over autonomous agents that they already expect from cloud workloads, rather than treating agent governance as a bolt-on after deployment.

Why it matters

AgentZ lands amid a broader wave of "agent operations" tooling — platforms that don't build new foundation models but instead manage how agents built on existing models are deployed, monitored, and constrained inside enterprises. As companies increasingly grant agents access to internal tools, credentials, and workflows, vendors with a security pedigree, like AccuKnox, are betting that the harder unsolved problem isn't building agents but safely operating many of them at once — a gap the company says is currently blocking agent rollouts from moving past internal pilots.