AI agents on Kubernetes
Run and Govern AI Agents in Kubernetes
kagent is the open source, Kubernetes-native runtime for AI agents, with identity, policy, and observability built in. agentregistry catalogs and governs the agents, MCP tools, and skills your teams are already building. Run agents from any framework with the same discipline you apply to the rest of your platform.
Open source and community-driven
One of the fastest-growing agent runtimes in cloud native
kagent was created at Solo.io and is now a CNCF Sandbox project, with contributions from the world’s leading technology companies.
Sandbox project
GitHub stars
Discord members
Contributors
Why teams choose kagent
Production discipline for every AI agent
Identity and policy
“Every agent needs a way to prove who it’s acting as, and what it’s allowed to do.”
kagent gives agents a workload identity with mTLS and fine-grained RBAC through Istio ambient mesh, policy-driven egress, and audit logs. Solo Enterprise for kagent adds agent identity that plugs into your existing OAuth and OIDC providers.
Governance
“Every team is spinning up agents, MCP tools, and skills with no central inventory or approval process.”
agentregistry gives you one open source registry to version, share, and discover every agent, MCP server, and skill. Solo Enterprise for agentregistry adds approval workflows, access control, and runtime policy.
Orchestration
“We can’t standardize on one agent framework, and scaling agents across clusters by hand doesn’t hold up.”
kagent runs agents built with Google ADK, LangGraph, or CrewAI as Kubernetes resources, connects them to tools over MCP and to each other over A2A, and scales them like any other workload.
Capabilities
Everything agents need to run in production
Declarative agents
Define agents, models, and tools as Kubernetes resources and manage them with the GitOps workflows you already use.
Bring any framework
Run agents built with Google ADK (Go or Python), LangGraph, or CrewAI side by side on one runtime.
MCP tools
Connect REST APIs, gRPC services, databases, and internal systems to agents through a declarative MCP spec.
Agent-to-agent (A2A)
Let agents discover, delegate to, and invoke each other across teams and clusters.
Built-in observability
OpenTelemetry traces, Prometheus metrics, and structured logs capture prompts, tool calls, and token usage.
Central registry
Catalog agents, MCP servers, skills, and prompts in agentregistry so teams can find and reuse what already exists.
Get hands-on
Try kagent in minutes
Self-paced labs and an on-demand workshop from the engineers who build kagent.
Trust and compliance
Enterprise ready from day one
SOC 2
Supports security and procurement review for enterprise and regulated environments.
Open source roots
kagent is a CNCF Sandbox project under Apache 2.0, and Solo.io contributed agentregistry to the CNCF in 2026.
Maintainer status
Solo.io created kagent and agentregistry, and its engineers maintain them, so fixes and features land upstream first.
Hardened releases
Enterprise distributions include long-term-support releases, SBOMs, and CVE scans.
Enterprise support
Solo Support with published SLAs for teams running agents in production.
Fits your stack
Works with the Kubernetes, Istio, OpenTelemetry, and Prometheus tooling your platform team already runs.
FAQ
Frequently asked questions
What is Solo.io’s platform for AI agents?
kagent is a Kubernetes-native runtime for building and running AI agents, adding identity, policy, tracing, and lifecycle management to the infrastructure you already run. agentregistry pairs with it as an open source registry to catalog and govern every agent, MCP server, and skill your teams deploy. Solo Enterprise for kagent and Solo Enterprise for agentregistry add enterprise features and support.
What agent frameworks does kagent support?
kagent is framework-agnostic. It runs agents built with Google ADK in Go or Python, LangGraph, and CrewAI, and connects them to any MCP tool. Agents can also talk to each other over the A2A protocol, so you don’t have to standardize on one framework to run agents in production.
How does kagent handle AI agent orchestration in Kubernetes?
Agents, models, and tools are Kubernetes resources, so kagent schedules, scales, and observes them with the same model your platform team already uses for the rest of your workloads.
What is agentregistry, and do I need one for agentic AI?
agentregistry is an open source registry for creating, versioning, sharing, and discovering AI agents, MCP servers, skills, and prompts. Once more than one team is shipping agentic AI, a registry keeps that sprawl from becoming unmanageable.
Is this open source, or do I need an enterprise contract?
Both projects are open source and free to run on your own. kagent is a CNCF Sandbox project, and Solo.io contributed agentregistry to the CNCF. When you need multi-cloud visibility, approval workflows, runtime policy, long-term-support releases, and SLAs, Solo Enterprise for kagent and Solo Enterprise for agentregistry add them on top.
See kagent run AI agents in your cluster
Request a demo to see how kagent and agentregistry bring identity, policy, and observability to the AI agents in your environment, or get hands-on first with a lab.
