HOL Guard

HOL Guard checks AI agent actions before they execute. Developers can run it locally without an account and connect it through supported agent hooks and managed MCP proxies. It reviews shell commands, file access, package installs and tool calls, then allows, blocks or requests approval according to policy. Coverage depends on the events exposed by each host.

Bitterbot's desktop extension and PraisonAI's tool-policy middleware example call the Guard CLI. This funding request would support maintenance of that public interface, regression tests for approval and failure behavior, and documentation other maintainers can use.

As of October 26, 2026, the public repository has 999 stars and 215 forks. Funded code and tests will remain Apache-2.0.