Comprehensive documentation for kagent — a Kubernetes-native framework for building, deploying, and managing AI agents.
| Document | Description |
|---|---|
| Overview | Architecture overview and core concepts |
| Human-in-the-Loop | Tool approval workflows and the Ask User tool |
| Git-Based Skills | Fetching agent skills from Git repositories |
| Agent Memory | Long-term vector-backed memory for agents |
| Prompt Templates | Reusable prompt fragments with Go templates |
| Multi-Runtime Support | Go and Python ADK runtime selection |
| Context Management | Event compaction for long conversations |
| Tools Ecosystem | Built-in tools: Kubernetes, Helm, Istio, and more |
| Agent CRD Reference | Full Custom Resource Definition reference |
| Document | Description |
|---|---|
| Enterprise Overview | Enterprise vs open source, architecture, Helm charts, pre-built agents |
| Security & Access Control | OIDC authentication, OBO tokens, AccessPolicy, RBAC, agentgateway |
| Multicluster Federation | Management-agent relay architecture, tunnel, cluster registration |
| Observability | Built-in OTel pipeline, distributed tracing, metrics, ClickHouse |
| Enterprise UI | Dashboard, agent chat, tracing visualization, inventory, user management |
| Agent Frameworks (BYO) | BYO agents with ADK, CrewAI, LangGraph |
| Model Providers | OpenAI, Anthropic, Azure, Gemini, Vertex AI, Bedrock, Ollama |
| Tools & MCP Servers | Built-in tools, K8s services as MCP, remote MCP, custom servers |
| Installation & Operations | Install, upgrade, debug, uninstall, licensing |
kagent is a CNCF project that enables development, deployment, and management of AI agents within Kubernetes environments. It provides:
- Kubernetes-native agent management via Custom Resource Definitions
- Multi-LLM support — OpenAI, Anthropic, Google Vertex AI, Azure OpenAI, AWS Bedrock, Ollama
- MCP tool integration — extensible tooling through the Model Context Protocol
- Human-in-the-loop — tool approval workflows and interactive user input
- Agent-to-Agent (A2A) protocol — multi-agent communication and coordination
- Long-term memory — vector-backed persistent memory with embedding search
- Declarative configuration — define agents, tools, and skills as YAML
- Observability — OpenTelemetry tracing and Prometheus metrics