Kagent
Cloud Native Agentic AI | Discord: https://bit.ly/kagentdiscord
Last verified:
What is Kagent?
kagent is an open-source programming framework that brings agentic AI to cloud-native environments, specifically built for DevOps and platform engineers. It enables AI agents to run directly in Kubernetes clusters to automate operations, troubleshoot issues, and solve complex cloud-native challenges. Created at Solo.io in 2025, kagent is a Cloud Native Computing Foundation (CNCF) sandbox project.
Unlike traditional chatbots, kagent leverages advanced reasoning and iterative planning capabilities to autonomously handle multi-step problems. It transforms AI insights into concrete actions, helping teams diagnose connectivity issues across multiple service hops, troubleshoot application performance degradation, automate alert generation from Prometheus metrics, debug Gateway and HTTPRoute configurations, and manage progressive rollouts with Argo Rollouts. The architecture consists of three main components: Tools (MCP-style functions for interacting with cloud-native systems), Agents (autonomous systems that plan, execute, and analyze tasks), and Framework (a flexible interface built on Google's ADK framework).
Key features include declarative agents defined as Kubernetes CRDs, multi-runtime support (Go and Python ADK), bring-your-own-frameworks (LangGraph, CrewAI, Google ADK), long-term persistent vector-backed memory, human-in-the-loop approval gates, agent-to-agent (A2A) communication, skills loaded from Git repos, prompt templates via ConfigMaps, context compaction with auto-summarization, agent sandboxing with RBAC, full observability with OpenTelemetry tracing and Prometheus metrics, and Postgres-backed storage. It supports multi-LLM providers including OpenAI, Anthropic, Google Vertex AI, Azure OpenAI, Ollama, and custom models.
kagent is designed for DevOps engineers, platform engineers, SREs, and operations teams working with Kubernetes who want to automate troubleshooting, incident response, observability, and platform self-service tasks using AI agents. Use cases include incident response agents that triage alerts, observability copilots for conversational PromQL, platform self-service for developers requesting resources, knowledge agents with RAG over runbooks, and multi-agent workflows for complex operations.
Kagent pricing
Pricing model: Freemium
kagent is completely free and open-source under Apache 2.0 license. The entire platform works with a single helm install with no add-ons, no extra databases, and no waiting for enterprise. All core features including agent lifecycle via CRDs, multi-runtime support, BYO frameworks, long-term memory, human-in-the-loop, agent-to-agent communication, skills from Git, prompt templates, context compaction, sandbox & security, full observability, and Postgres storage are included in the open-source version. Solo Enterprise for kagent is available as a paid enterprise distribution with custom pricing (median buyer pays $68,000/year according to Vendr data) that adds advanced agent management interface, observability tools, and multi-cluster federation support across federated Kubernetes clusters. Pricing at Solo.io depends on environment and use case with no single price list - customers must engage with a specialist for tailored estimates.
Kagent pros
- Open-source with Apache 2.0 license
- CNCF Sandbox project with public roadmap
- Kubernetes-native agents as CRDs for GitOps workflow
- No add-ons or extra databases required - single helm install
- Multi-runtime support for Go and Python ADK
- Bring your own framework - LangGraph, CrewAI, Google ADK supported
- Works with any LLM - OpenAI, Anthropic, Gemini, xAI, Azure, AWS Bedrock, Ollama
- Native Model Context Protocol (MCP) support for tool integration
- Agent-to-Agent (A2A) communication with first-class delegation
- Persistent vector-backed long-term memory across sessions
- Human-in-the-loop with tool approval gates and agent-initiated questions
- Built-in observability with OpenTelemetry tracing and Prometheus metrics
- Postgres-backed storage with reviewable migrations - no proprietary lock-in
- Agent sandboxing and RBAC security hardening out of the box
- Skills loaded from Git repos - agents learn your runbooks and ADRs
- Context compaction with auto-summarization to stay within token budgets
- Prompt templates stored as ConfigMaps for DRY system prompts
- 2,500+ GitHub stars and 2,800+ Discord members active community
- Built by the founders of Istio with production-grade reliability
- Zero-trust ready with Istio or Ambient Mesh support and mTLS
Kagent cons
- Requires OpenAI API key or other LLM provider API key to run agents
- Needs kind, Helm, and kubectl installed before getting started
- Kubernetes cluster required - not suitable for non-Kubernetes environments
- Learning curve for teams unfamiliar with Kubernetes CRDs and GitOps
- Dependent on external LLM provider costs (OpenAI, Anthropic, etc.)
- Limited documentation compared to more mature frameworks
- Relatively new project (created in 2025) with less real-world adoption
- Enterprise features require Solo Enterprise purchase with custom pricing
- UI requires port-forwarding to access dashboard locally
- Complex setup for development mode with separate frontend/backend/controller
Frequently asked questions about Kagent
What's the best way to get started with kagent?
The best way to get started with kagent is to follow the quickstart guide. This will give you a basic understanding of how kagent works and how to use it. Before beginning, ensure you have kind for creating a local Kubernetes cluster, Helm for installing the kagent chart, and kubectl for interacting with your cluster installed. You'll also need an OpenAI API key. Then download the kagent CLI using brew install kagent or the curl script, and install kagent to your cluster with kagent install --profile demo for a demo profile with preloaded agents and MCP tools.
What differentiates kagent from other LLM frameworks?
kagent is special for three core reasons: Declarative - kagent is designed from the ground up to be declarative where you define agents, tools, and instructions and kagent takes care of the rest, while most other frameworks are procedural requiring code to tell the LLM what to do. Kubernetes Native - kagent is designed for Kubernetes environments and integrates seamlessly with services already running in the cluster. Easy To Use - kagent abstracts all the hard parts of building an agent so you can focus on connecting your business logic to the agent.
How do I report bugs or request features?
The best way to report bugs or request features is to create an issue on the GitHub repository at github.com/kagent-dev/kagent. The project maintains a public roadmap and public RFCs, and PRs are welcome.
How do I contribute to the kagent project?
The best way to contribute is to check out the contribution guide and submit a PR. The project has 100+ contributors ranging from individual operators to teams at hyperscalers, with weekly releases and monthly community demos. Maintainers hang out in Discord and answer in hours, not days.
What LLM providers does kagent support?
kagent supports multi-LLM providers including OpenAI, Anthropic, xAI, Google Gemini, Azure OpenAI, AWS Bedrock, Vertex AI, Ollama, and Hugging Face. There is no lock-in at any layer of the stack, and you can bring any LLM with native MCP, A2A, and OpenAI-compatible endpoints.
What agent frameworks can I use with kagent?
kagent supports bring-your-own-frameworks including LangGraph, CrewAI, Google ADK, OpenClaw, NemoClaw, OpenShell, Hermes Agent, MCP, A2A, Python, and Go. kagent orchestrates any agent framework you bring without requiring rewrites.
Is kagent production-ready?
Yes, kagent is production-grade. It is a CNCF Sandbox project with production-grade Postgres-backed storage, agent sandboxing, RBAC, and security hardening out of the box. It has full observability with OpenTelemetry tracing, Prometheus metrics, and structured logs. The project has 2,500+ GitHub stars, 2,800+ Discord members, and 100+ contributors including teams at hyperscalers.
What use cases does kagent support?
kagent supports incident response (pager-aware agents that triage alerts and draft runbooks), observability copilot (conversational PromQL and trace navigation), platform self-service (developers request namespaces/databases/pipelines in plain English), knowledge agents (RAG over runbooks, ADRs, and Slack history with mTLS and RBAC), chat providers (messaging agents through Telegram, WhatsApp, Slack, or Discord), and multi-agent workflows (chaining agents to automate complex operations end-to-end).
How does kagent handle agent memory?
kagent provides long-term persistent vector-backed memory across sessions. Agents remember context, not just the last prompt. This persistent memory allows agents to maintain continuity across conversations and sessions.
What security features does kagent include?
kagent includes agent sandboxing, RBAC, and security hardening out of the box for running untrusted code safely. It is zero-trust ready with support for Istio or Ambient Mesh, mTLS, fine-grained RBAC, and policy-driven egress for agent traffic. It also includes built-in security and privacy guardrails via NVIDIA NemoClaw with policy enforcement on every LLM call.