kubeswarm FAQ - Orchestrate Agents on Kubernetes

Do I need a cloud LLM API key?

Frequently asked questions about kubeswarm - the Kubernetes operator for agent orchestration.

What models are supported?

Any model accessible via Anthropic, OpenAI, Google Gemini APIs, or any OpenAI-compatible endpoint (Ollama, vLLM, LM Studio). See LLM Providers.

Does kubeswarm run the LLM?

No. kubeswarm orchestrates agent pods that call external LLM APIs. It does not host or serve models. The LLM runs wherever you point it - cloud API, self-hosted Ollama, vLLM, etc.

What is the MCP gateway?

An operator-managed SSE server that exposes SwarmAgent capabilities as MCP tools. This enables agent-to-agent calls and external MCP client access. It is an implementation detail - you don't need to configure it directly.

How do I manage API keys?

Use native Kubernetes Secrets. Reference them via spec.apiKeyRef (single key) or spec.envFrom (multiple keys). See API Key Management.

Is there a custom secret CRD?

No. kubeswarm uses native Kubernetes Secrets only. No wrapper CRD.

How does budget enforcement work?

The operator tracks rolling 24h token usage per agent. When spec.guardrails.limits.dailyTokens is exceeded, replicas are scaled to 0 and a BudgetExceeded condition is set. Replicas restore automatically when the window rotates. See Budget Management.

Can agents call other agents?

Yes, via the spec.agents[] section. See Agent-to-Agent.

What happens if an MCP server goes down?

The operator probes MCP servers every 60s. If unreachable, the MCPDegraded condition is set and (if configured) a SwarmNotify alert fires. The agent continues running - it just can't use tools from that server.

How do I contribute?

See the GitHub repo. Issues and discussions welcome.