kagent takes a Kubernetes-first approach to AI agent deployment by running agents as native Kubernetes resources rather than external applications that connect to clusters through API proxies. The framework defines custom resource definitions for agent configurations, tool bindings, and execution policies, letting platform engineers manage AI agents with the same GitOps workflows they use for other infrastructure components. The kmcp toolkit provides Kubernetes-specific tools for common operations like pod inspection, log analysis, resource scaling, and configuration management.
The architecture separates agent logic from infrastructure concerns through a clean abstraction layer. Agents receive structured context about cluster state and can invoke Kubernetes operations through approved tool interfaces, reducing the risk of uncontrolled changes to production infrastructure. The CNCF sandbox acceptance signals community validation of the project approach and provides a governance framework for contributions from organizations building their own platform engineering capabilities.
Created at Solo.io, kagent addresses the growing demand for AI-assisted Kubernetes operations without requiring teams to build custom agent integrations from scratch. With over 2,500 GitHub stars and active development, it fills a specific gap between general agent frameworks that lack Kubernetes awareness and existing Kubernetes tools that lack AI reasoning capabilities. The Apache 2.0 license enables commercial adoption, and the project documentation includes patterns for common platform engineering scenarios.