DashClaw

DashClaw is an approval layer for AI agents that run unattended. Policies hold risky agent actions, such as a production deploy or a destructive database command, before they execute, and a person approves or denies them later from a dashboard or a phone while routine work continues. Every decision is written to a signed ledger that records the declared goal, risk score, matched policy, approver and reported outcome. It is aimed at developers and teams who leave coding and operations agents running without supervision. The vendor frames it as a complement to the permission prompts in Claude Code and Codex: those protect a user sitting at the keyboard, while DashClaw freezes a dangerous call and lets someone approve it minutes or hours afterwards from anywhere. Enforcement happens at the hook layer in Claude Code, Codex and Hermes and at the OpenClaw gateway, and the Node and Python SDKs, an MCP server and a REST API support cooperative checks. A single command installs the runtime, provisions Postgres, creates a key and wires the first hook, and approvals can be routed to channels such as Discord. Pricing is published. Self hosted is free under the MIT license, with the core guard, policy, approval and ledger features, no telemetry, and no seat or action limits. Hosted Indie costs $49 per month for 2 seats, 30 days of retention and email support. Hosted Team costs $199 per month for 10 seats, 90 days of retention, organization scoped rate limits and priority support. Hosted plans carry monthly action ceilings documented separately, and a 30 day hosted trial needs no card. Hosted workspaces can export their full decision history and configuration as one file and import it into a self hosted instance.