Claude Code¶
Claude Code is Anthropic's command-line coding agent. It reads files, writes code, runs shell commands, manages git, and spawns its own subagents — an autonomous developer that lives in your terminal. On the stack-sovereignty-box it is the on-box "VPS architect": the agent that actually installs software and changes system state, tasked by hermes when something must happen on the box.
Why it matters for owned AI infrastructure¶
Claude Code turns natural-language intent into real changes on real infrastructure. Paired with hermes (the always-on orchestrator), it forms a two-layer agent stack: Hermes decides and delegates; Claude Code executes on the box. This is the concrete shape of "run your own AI agents on hardware you own."
Key facts¶
- CLI agent (
claude); print mode (claude -p) runs one-shot tasks non-interactively — ideal for automation and orchestration. - Interactive TUI mode for multi-turn work; supports slash commands, checkpoints, plan mode.
- Speaks model-context-protocol (MCP) to connect external tools/data.
--max-turns/--max-budget-usdguardrails;--allowedToolsrestricts capabilities.- Extensible via plugins (e.g. the
vps-architectplugin that seeded this box's skills) andCLAUDE.mdproject memory.
Related¶
- hermes — the orchestrator that tasks Claude Code
- model-context-protocol — how it connects to tools
- ai-agent-orchestration — the multi-agent pattern it participates in