Finding NemoClaw, Solo fishes out production agentic runtime for Kubernetes

No-code declarative agents

Programmatic agents built with frameworks like Google ADK, LangGraph and CrewAI run side-by-side with no-code declarative agents on a kagent runtime providing consistent and uniform end-to-end auditability, on-behalf-of policy enforcement and human-in-the-loop controls. 

In less than a year as a CNCF Sandbox project, kagent has attracted more than 300 contributors from organisations including Microsoft, Amazon, Oracle, Amdocs and Orange. 

While programmatic and declarative agents cover most of enterprise agentic use cases today, a third pattern is rapidly emerging as the next mode for any production agent runtime: the OpenClaw-style always-on assistant. 

These agents run persistently, retain context across days or weeks and take long-running action on behalf of their users. 

Finding NemoClaw

Capitalisation enthusiasts at Nvidia have said that Nvidia’s NemoClaw, built on the open source OpenClaw project, has emerged as the reference implementation of this pattern, with a sandbox-first foundation providing safety and isolation controls for assistant agent runtimes. 

That foundation is well-suited to individuals and small teams running an assistant on a single host, but enterprises need the same pattern operating across hundreds of assistants under unified identity, policy and observability. Adding direct support for NemoClaw in kagent scales this always-on assistant pattern from a strong reference deployment into a fleet-scale production runtime. 

“At Nvidia GTC, Jensen Huang said: ‘Mac and Windows are the operating systems for the personal computer. OpenClaw is the operating system for personal AI,'” said Idit Levine, founder and CEO of Solo.io. “Our customers and the kagent community have been driving us to the natural next step: extending that operating system beyond the personal device into the distributed Kubernetes infrastructure where their business actually runs. With NemoClaw in kagent, that’s exactly what we’re delivering.”

Kagent complements and extends NemoClaw to deliver an enterprise governance layer for always-on assistants with capabilities platform teams need to operate fleets of assistants in production, including:

• Security and isolation: Assistants act under each user’s identity and policies with on-behalf-of authorisation, allowing the scope and authority of an agent to be constrained based on context and use case.
• End-to-end Observability. Built-in telemetry and tracing show each step in the agentic loop of an assistant, showing exactly what and when an agent takes action.
• Multi-cluster and federation. A central, multi-cluster control plane governs assistants running across regions and clusters, while providing the ability to schedule and execute agents in specific environments.
• Governance and lifecycle. Declarative config, GitOps workflows and progressive rollouts let teams ship and update assistant blueprints across many clusters using the same process they use for application code.

The Solo.io agentic stack provides everything teams need to develop, deploy and operate AI agents on Kubernetes and is powered by core open source projects created and contributed to the community by Solo.io including agentgateway, a Linux Foundation project, is the high-performance data plane for agentic traffic, built from the ground up in Rust to span the full range of agentic protocols and use cases as a unified inference gateway, LLM gateway, MCP gateway and A2A gateway.