kagent runs AI agents as Kubernetes custom resources

A new technical guide demonstrates how to use kagent, a CNCF Sandbox project, to run AI agents as Kubernetes resources. By treating agents as custom resources, users can define them using system prompts, models, and tools within standard YAML manifests. The guide specifically shows how to build an 'agent builder' that uses a local model via Ollama to generate and deploy other agents.

Managing agents via Kubernetes CRDs

• Agents are implemented as Custom Resources in kagent.
• The controller converts an Agent manifest into a running pod that communicates via A2A.
• The project supports local models through Ollama or hosted providers like OpenAI, Anthropic, and Gemini.
• A tool server provides Kubernetes-related capabilities to the agents.
• The guide demonstrates a restricted RBAC configuration where a builder agent is only permitted to create agents.kagent.dev resources.

Why it matters

By integrating AI agents directly into the Kubernetes resource model, kagent brings agent lifecycle management into existing cloud-native workflows.