Kaigents
Get Started
← All articles
sovereign AIKubernetesopen sourcegetting started

What is sovereign AI compute, and why does it matter?

Kaigents

Sovereign AI compute is the idea that you should own the infrastructure your AI runs on. Not rent it from a cloud provider by the token. Own it.

The difference is control. When you use a hosted AI API, you pay per request. Your data leaves your network. You are subject to someone else's rate limits, pricing changes, and uptime. When you run your own models on your own hardware, none of that applies.

Why now?

Three things changed:

  1. Open-weights models are good enough. Qwen3, Llama, and Mistral can handle real production workloads. You do not need GPT-4 to write a blog post or summarize a document.

  2. Commodity hardware can run them. AMD Ryzen AI processors and consumer GPUs can serve 30-70 billion parameter models. You do not need a data center.

  3. Kubernetes makes it manageable. If you already run a cluster, deploying an AI agent platform is no different from deploying any other workload. CRDs, controllers, GitOps — the tools you already use.

What you get with Kaigents

Kaigents is a Kubernetes-native AI agent platform. That means:

  • Agents are CRDs. You define an agent in YAML, apply it with kubectl, and the controller handles the rest. No separate deployment pipeline.

  • Skills constrain output. A skill is a process-adherence playbook. Instead of letting an agent free-form generate, you define the steps, quality gates, and expected output format. The agent follows your methodology.

  • Temporal handles durability. Long-running agent tasks survive restarts, network failures, and cluster disruptions. If a node goes down, the task picks up where it left off.

  • No per-token costs. You run models on your own hardware. The only cost is electricity and hardware depreciation.

What you give up

Sovereign compute is not the right choice for every workload:

  • You need hardware. A machine that can run a 30B model costs money up front. If your workload is sporadic, a cloud API may be cheaper.

  • You manage infrastructure. Kubernetes, model servers, storage — you operate it. If you do not have a Kubernetes cluster, there is a learning curve.

  • You pick models. No one is routing your requests to the best available model. You choose which open-weights model to serve, and you live with that choice until you change it.

When sovereign compute wins

Sovereign AI compute is the right choice when:

  • You run enough AI workload that per-token costs are significant
  • Your data cannot leave your network (compliance, privacy, competitive advantage)
  • You already operate Kubernetes and want to add AI capabilities
  • You want predictable costs instead of usage-based billing
  • You want to build AI agents that do real work, not just chat

Start exploring

If you have a Kubernetes cluster and want to see how Kaigents works, start with the getting started guide.

If you want to understand the architecture first, read how it works.