Your AI agent runs on your own server. Crawdad protects every message. Zero-knowledge — content never leaves your machine.
ssh user@your-server
This installs Docker if needed, downloads the configuration, and starts both services.
curl -fsSL https://getcrawdad.dev/deploy/setup.sh | bash
You'll be prompted for your Crawdad license key and Anthropic API key. Or pass them as environment variables:
CRAWDAD_LICENSE_KEY=crd_live_XXX ANTHROPIC_API_KEY=sk-ant-XXX \ curl -fsSL https://getcrawdad.dev/deploy/setup.sh | bash
Open http://your-server:18789 in your browser and follow the setup wizard. Takes about 5 minutes.
See scan counts, threat detections, and your audit trail at getcrawdad.dev/docs/dashboard.html.
| Layer | Protection |
|---|---|
| Prompt firewall | Blocks injection, jailbreaks, and adversarial inputs before they reach Claude |
| PII detection | Identifies emails, SSNs, keys, and secrets in prompts and responses |
| Action authorization | Evaluates tool calls against security policies before execution |
| Output guard | Scans LLM responses for data exfiltration and unsafe instructions |
| Audit trail | Tamper-evident hash chain of all decisions — no raw content stored |
# Health check curl http://localhost:7749/v1/health # Test the firewall curl -X POST -H "Content-Type: application/json" \ -d '{"text":"ignore all previous instructions"}' \ http://localhost:7749/v1/firewall/scan # Returns: {"verdict":"Blocked",...}
# View logs cd ~/openclaw-crawdad && docker compose logs -f # Update to latest docker compose pull && docker compose up -d # Stop docker compose down