Kubernetes-native Agent Platform for Teams and Enterprise

Demo

See KubeOpenCode in action.

Features

Concrete capabilities shipped in recent releases.

⚡

Live Agents

Agents run as persistent Deployments with zero cold start. Shared context across tasks, session history survives restarts.

🌐

Web Terminal

Access any Agent directly from the browser via the built-in web terminal. Real-time interactive sessions without SSH or port-forwarding.

💤

Standby (Auto Suspend/Resume)

Agents auto-suspend after idle timeout, auto-resume when new Tasks arrive. Connection-aware — active sessions prevent suspension.

⏰

CronTask

Scheduled and recurring task execution. Concurrency policies (Allow/Forbid/Replace), manual trigger support, and retention control.

🔄

Git Auto-Sync

HotReload syncs Git contexts in-place without restart. Rollout policy triggers rolling updates with active Task protection.

📊

Concurrency & Quota

Limit concurrent tasks per Agent and rate-limit task starts with sliding time windows. Tasks queue automatically when limits are reached.

🧩

Skills

Reusable AI agent capabilities from Git repos. Share skills across Agents and templates, auto-injected as slash commands.

☸️

Declarative CRDs

Fully Kubernetes-native. GitOps-friendly, works with Helm, Kustomize, and ArgoCD. Just kubectl apply.

How It Works

Two simple resources. One powerful workflow.

Define an Agent

Deploy persistent AI agents your team can interact with in real time — through the web terminal, CLI, or by submitting Tasks.

Zero cold start — agent is always running
Interactive terminal access via CLI or web
Auto-suspend when idle, resume on demand
Session history persists across restarts

agent.yaml
apiVersion: kubeopencode.io/v1alpha1
kind: Agent
metadata:
  name: dev-agent
spec:
  profile: "Interactive development agent"
  workspaceDir: /workspace
  port: 4096
  persistence:
    sessions:
      size: "2Gi"
  standby:
    idleTimeout: "30m"

Submit a Task

Run stable, repeatable AI tasks in ephemeral Pods. Perfect for CI/CD pipelines, batch operations, and automated workflows.

No new tools to learn — just kubectl apply
Works with any CI/CD pipeline
Scale with Helm templates for batch operations
Rate limiting and quota controls

task.yaml
apiVersion: kubeopencode.io/v1alpha1
kind: Task
metadata:
  name: update-dependencies
spec:
  templateRef:
    name: ci-runner
  description: |
    Update all dependencies to latest versions.
    Run tests and create a pull request.

Architecture

A simple, Kubernetes-native design with no external dependencies.

Task

WHAT to do

Agent

HOW to execute

Pod (OpenCode)

WHERE it runs

No external dependenciesNo PostgreSQL, no Redis — just Kubernetes (etcd for state, Pods for execution)

Single container imageController, init containers, and utilities all in one image

Two-container patternInit container copies OpenCode binary, worker container runs the agent

Frequently Asked Questions

Yes. KubeOpenCode runs on any standard Kubernetes cluster (v1.26+). You can use managed services like EKS, GKE, AKS, or a local cluster with Kind or minikube for development.

KubeOpenCode uses OpenCode as its AI engine, so it supports every provider and model that OpenCode supports — including Anthropic, OpenAI, Google Gemini, Amazon Bedrock, Azure, Google Vertex, xAI, Mistral, Groq, OpenRouter, GitHub Copilot, and more. You can also use any OpenAI-compatible endpoint. See the full list of 75+ supported providers.

KubeOpenCode itself is free and open-source (Apache 2.0). You pay for your own infrastructure (Kubernetes cluster) and any AI model API keys you choose to use.

KubeOpenCode is currently in early alpha (v0.1.x). The API may change without backward compatibility. We recommend using it for development, testing, and evaluation while we work toward a stable release.

Ready to run AI agents on Kubernetes?

Get started in minutes with Helm. Deploy your first agent today.

Get Started View on GitHub