Download
Kagent is a Kubernetes-native framework for building, deploying, and operating AI agents as first-class cloud-native workloads. It models core agent concepts declaratively using Kubernetes custom resources, so teams can manage agents similarly to other platform components via YAML, controllers, and standard cluster workflows. In kagent’s design, an “Agent” represents a system prompt plus a set of tools and other agents, along with an LLM configuration, making the agent definition portable and repeatable across environments. It supports multiple model providers through a dedicated configuration resource, allowing teams to switch providers or run mixed environments while keeping the agent spec stable. A major focus is tool integration via MCP: agents can connect to MCP servers for tool access, and kagent includes an MCP server with tools for common Kubernetes and platform engineering systems.
Features
- Kubernetes-native agent framework built around custom resources and controllers
- Declarative Agent definitions that bundle prompts, tools, and model configuration
- Multi-provider LLM support via a dedicated model configuration resource
- MCP-based tooling with built-in integrations for Kubernetes and platform stacks
- OpenTelemetry tracing support for observability into agent and tool activity
- Multiple operational surfaces including UI and CLI for managing agents
Project Samples
