Security / Local-First Control
SmallClaw Security: Local-First AI Agent Control on macOS
SmallClaw is designed as a local-first Mac application. It uses local permissions, approval gates, audit logs, Keychain-backed secrets, and encrypted local persistence to keep AI execution bounded by user-controlled policy.
Controlled Capabilities
- Shell execution, file deletion, and outbound actions are governed by explicit settings and approval rules.
- Browser automation and desktop automation require appropriate macOS permissions and user-controlled policy.
- Audit logs make sensitive agent activity easier to review after execution.
- Model providers and endpoints are configurable by the user.
Limits
SmallClaw should not be described as unrestricted automation or as protection against a fully compromised device. It is a local-first AI agent app that gives users policy controls, approvals, and reviewable records around execution.
FAQ
Does SmallClaw require approval for risky actions?
SmallClaw includes permission settings and approval gates for sensitive capabilities such as shell execution, file deletion, outbound actions, browser automation, desktop automation, and other higher-risk operations.
Does SmallClaw run 24/7 when the app is closed?
SmallClaw is a macOS desktop application. Scheduled runs, channel polling, and automation depend on the app being available to run. It should not be described as a strict always-on 24/7 daemon when the app is closed.