Kagent

Kagent is an open source framework that runs AI agents inside Kubernetes clusters. It treats agents as another workload on the cluster, so platform and DevOps teams can deploy, observe and govern them with the tools they already use. The project is a CNCF Sandbox project, and its site states it was built by the founders of Istio. Agents are defined declaratively as Kubernetes custom resources, versioned in Git and rolled out with kubectl and GitOps workflows. The framework supports Go and Python agent runtimes, works with any LLM provider including OpenAI, Anthropic, Gemini and xAI, and can orchestrate agents built with LangGraph, CrewAI or Google ADK. It supports MCP tool servers, the A2A protocol and OpenAI compatible endpoints. Listed capabilities include persistent vector backed memory across sessions, human approval gates on tool calls, OpenTelemetry traces, Prometheus metrics and structured logs, and mTLS and RBAC when run on Istio or Ambient Mesh. An Agent Substrate runtime and NVIDIA NemoClaw guardrails are also listed. Installation uses a one line script or Homebrew and requires kind, Helm and kubectl. The vendor states every feature works from a single Helm install with no add ons or enterprise tier. The site publishes no pricing page, and the project is presented as open source with a Get Started and GitHub path.