Our verdict
Local-first runtime control for AI coding agents (Claude Code, Codex, Cursor and others). It reviews supported shell commands, secret reads, package installs, MCP changes and related actions against your policy, then can allow, ask or block before they run, with a local receipt trail. The core Guard Local package is free and open source under Apache-2.0 and works offline; optional Guard Cloud adds sync, longer history and team policy. Useful if you let agents act on a real machine and want a second checkpoint beyond each tool's native approvals.
You run AI coding agents on a real machine and want portable pre-action checks, local receipts and optional team policy on top of each agent's own approvals.
You only use agents in a locked-down sandbox, you need a guaranteed prompt-injection preventer, or you will not install a local Python/CLI security layer.
Pricing: Guard Local free forever (Apache-2.0, no account). Guard Cloud: Solo $4.99/mo, Pro $15/mo, Team $30/seat/mo; Enterprise contact sales (check site). Annual billing offered with two months free on the pricing page.