Self-host DashClaw

DashClaw is a governance control plane for unattended AI agents. It evaluates agent tool calls against declarative policies before execution, either allowing, warning, containing, requiring human approval, or blocking them. Mechanical enforcement works at the hook seam for several coding agents (fail-closed, exit-2 on block), while SDKs, MCP server, and REST API integrations are cooperative — the caller must honor the verdict. Approvals can be resolved from a web dashboard, CLI, mobile PWA, Telegram, or Discord. The system includes a calibrated interruption controller that learns from approve/deny verdicts to reduce false interruptions with a proven bound, an enforcement liveness probe that tests whether the installed hook seam actually holds actions, prompt-injection scanning on by default, and Ed25519-signed receipts for decision records. It runs on Node 20+ with PostgreSQL (Docker or Neon serverless), defaults to port 3000, and can be deployed locally via npx dashclaw up or to Vercel with a Neon database. The project is MIT-licensed, maintained by an AI agent under a human-held charter (Practical Systems), and has near-daily releases since February 2026.