Back
D
runtime-guard · self-host New · 2026-08-06

DashClaw

A self-hosted, MIT-licensed approval layer and runtime guard for unattended AI agents. It intercepts agent actions (such as shell commands, file edits, or deployments) before execution, evaluates them against safety policies, and freezes risky actions until human approval is provided.

Visit site Open source via hn-showLast verified ·

AFIT Review

Accuracy
n/a
Privacy
5.0
Value
5.0
Multilingual
n/a
Ease of use
4.0
Scalability
n/a

Lab Verdict

Ideal for developers running unattended AI coding agents who need a reliable, fail-closed safety gate to prevent destructive actions on their infrastructure. It is easy to spin up locally with a single command and keeps all data self-hosted. However, mechanical enforcement is currently restricted to specific agents like Claude Code, Codex, and Hermes, while other environments must cooperate via SDKs or MCP.

llm-judge · website review · 2026-08-07

Pricing

No pricing captured.

Capabilities

agent-governanceguardrailsagent-safetymcp-server