DashClaw: Decision Infrastructure for AI Agents

DashClaw is an open-source GitHub project providing decision infrastructure for AI agents. It enables interception of agent actions, enforcement of guard policies, human approval workflows, and the creation of audit-ready decision trails. The project supports various agent frameworks and runtimes, including LangChain, CrewAI, AutoGen, and OpenAI Agents SDK.

RDR85Confidence 90%AI agentsgovernancesafetyauditdeveloper toolsopen sourceLangChainCrewAIAutoGen

Why it matters

As AI agents become more autonomous and interact with real-world systems, robust governance and safety mechanisms are crucial. DashClaw addresses this by providing a centralized layer to control agent actions, ensure compliance with policies, and maintain an auditable record of decisions. This helps mitigate risks associated with agent autonomy and facilitates responsible AI deployment.

DashClaw is a GitHub project designed to serve as a governance layer for AI agents. Its core functionality revolves around intercepting actions initiated by AI agents, allowing for policy enforcement, human approval, and comprehensive auditing. The system is built to sit between agents and external systems, evaluating each potentially risky action before execution.

Key features include the ability to block, warn, or hold actions for approval based on defined policies. It supports identity verification for agents using JWKS-verified OIDC bearer tokens, offering cryptographic attribution and replay protection. Declarative policies can be implemented to manage risk thresholds, deployment gates, capability access, and semantic checks. For actions requiring human oversight, DashClaw routes approval requests to a dashboard, CLI inbox, mobile PWA, Telegram, or Discord.

Every agent action processed by DashClaw generates a replayable decision record, detailing the agent's goal, reasoning, risk score, matched policies, assumptions, and evidence. The system also ensures that terminal outcomes are durable and prevents double execution from retried agents. Furthermore, it offers a capability registry to wrap HTTP APIs with per-agent access rules, rate limits, and audit trails. DashClaw integrates with various agent runtimes and frameworks, such as Claude Code, Codex, Hermes Agent, OpenClaw, LangChain, CrewAI, AutoGen, LangGraph, and OpenAI Agents SDK.