One gate, every surface you already use.
Mergen hooks into the agent or IDE you're already running, with nothing to run beside it and no new place to watch. Pick yours below.
Put a deterministic gate on Claude Code: every AI agent tool call and every Bash command is intercepted before it executes. Block destructive actions, hold risky ones for Slack approval.
Learn more →Put the same deterministic gate you run for Claude Code in front of OpenAI's Codex CLI: every shell command and every apply_patch edit is evaluated before it runs.
Learn more →Gate Cursor’s agent mode with a local execution gate: shell commands are checked against deterministic policy before they run, and file edits are checked as they land and reverted if they break policy.
Learn more →Add an enforcement layer to Windsurf’s Cascade agent: a local execution gate that blocks destructive tool calls in milliseconds and holds risky ones for Slack approval.
Learn more →Check every file edit an AI agent makes in VS Code against deterministic detectors, and revert one that breaks policy right after it lands. Works under any agent: Copilot, Cline, and others.
Learn more →Slack · PagerDuty: outbound alerts and inbound webhooks, not a tool-call gate.
In a terminal, mergen shell-init gates what you type (a pre-execution block in zsh enforce mode; a warning in bash, fish and zsh advise mode). In a CI job or script, mergen-server exec -- <cmd> runs the command through the same gate first.