Show HN: OkaiDokai, tool-level firewall for OpenClaw, Claude Code and Codex (okaidokai.com)

OkaiDokai has launched a new tool-level firewall designed to enhance the security of AI agents like OpenClaw, Claude Code, and Codex by intercepting all tool calls—including shell commands, HTTP requests, and file writes—in real-time. This innovative system allows users to approve or deny these calls through push notifications on various devices, including iOS, Android, and watchOS. Each decision can contribute to an evolving ruleset, making OkaiDokai smarter over time, similar to how spam filters learn. The significance of OkaiDokai lies in its application of immediate user control over potentially destructive actions performed by AI agents, with decisions processed in under 1 millisecond. The tool not only logs every interaction—providing a detailed audit trail of actions approved or denied—but also incorporates features like glob-based pattern matching and customizable rulesets. This level of monitoring and user engagement could greatly improve security protocols in AI/ML environments, addressing concerns about unregulated AI behavior and enhancing accountability. With seamless integration and minimal setup requirements, OkaiDokai stands out as a vital addition to the AI toolkit landscape.