DashClaw

DashClaw is an open-source decision infrastructure platform for AI agents that sits between your agents and external systems, evaluating policies before any action executes and recording verifiable evidence of every decision. It provides control before execution — not just observability after the fact — making it suitable for teams that need governance, compliance, and human-in-the-loop oversight for autonomous AI workflows. DashClaw supports LangChain, CrewAI, OpenAI, Anthropic, AutoGen, Claude Code, Codex, Gemini CLI, and custom agents via SDK, MCP server, Claude Code hooks, or a zero-code skill install.

• Guard Policies: Declarative rules that intercept risky agent actions before they execute, with configurable risk thresholds, auto-allow, warn, and require-approval modes.
• Human-in-the-Loop (HITL) Approvals: Approval queue with risk scores and one-click Allow/Deny via dashboard, CLI, mobile PWA, or Telegram bot.
• MCP Server Integration: @dashclaw/mcp-server exposes 8 governance tools and 4 resources over Model Context Protocol for zero-code integration with Claude Code, Claude Desktop, and Managed Agents.
• SDK Support (Node.js & Python): Install via npm install dashclaw or pip install dashclaw and wrap agent actions in the 4-step Guard → Record → Verify → Outcome loop.
• Claude Code Hooks: Drop-in hooks govern 40+ tool types with semantic classification, risk scoring, and token usage capture — no SDK instrumentation needed.
• Drift Detection: Statistical behavioral drift analysis surfaces critical alerts when agents deviate from established baselines.
• Capability Registry: Full CRUD, HTTP invocation, health monitoring, and per-agent access rules for governing external API calls.
• Workflow Engine: Multi-step workflows with governance at every step, variable substitution, continue_on_failure, and resume-from-checkpoint support.
• Analytics Dashboard: Cost trends, action volume, agent and type breakdowns, policy enforcement stats, and token efficiency with 7d/30d/90d time ranges.
• Prompt Injection Scanning: On by default — detects and blocks injection patterns in declared agent goals.
• CLI Approval Channel: @dashclaw/cli lets developers approve or deny pending agent actions directly from the terminal.
• Self-Hostable on Vercel + Neon: Deploy for $0 using Vercel free tier and Neon free tier with one-click setup and automatic database schema creation.