Quick Start

Requires Node.js >= 18. No account needed.

Install

npm install -g guardclaw
guardclaw start

Takes about 30 seconds. First launch opens an interactive setup wizard.

Setup Wizard

The wizard walks you through four choices:

1. Evaluation mode — local, mixed, or cloud
2. LLM backend — pick a local model (LM Studio, Ollama, built-in MLX) or a cloud provider (Claude, OpenAI Codex, MiniMax, Kimi, OpenRouter, Gemini)
3. Response mode — Auto (flag and block risky calls) or Monitor only (log without blocking)
4. Agent connections — auto-detects installed agents and installs hooks/plugins with one confirm

Re-run any time with:

guardclaw setup

Restart your AI agent after installing hooks.

Dashboard

Once running, open the dashboard at localhost:3002.

The dashboard shows:

• Live event timeline with risk scores
• Per-session tool call history
• Blocking toggle (on/off without restart)
• Settings panel for LLM backend configuration

Risk Tiers

Score	Verdict	Behavior
1–3	SAFE	Runs normally
4–7	WARNING	Logged with elevated audit signal
8–10	HIGH RISK	Blocked / requires approval

Uninstall

guardclaw stop
npm uninstall -g guardclaw
node scripts/install-claude-code.js --uninstall  # remove hooks