Every tool call your agent makes is checked before it runs, on your computer or on our server. Safe work goes straight through, risky steps wait for your OK, and the damaging ones never happen.
The check takes 9 ms at p95. Your agent doesn't feel it.
Tell us a little about you, then pick where it runs: on your computer, or on our server with a dashboard for your team.
Already signed up? Use the setup link from last time, or write to hello@squidbrake.com.
We see your name, email, the agents you picked, and for each action that was blocked or sent for approval:
the program (e.g. rm), the rule and what happened. Never your commands, files or prompts.
Name, company, email. No password, no card.
On your computer (one line, Windows or Mac), or on our server with a dashboard for your team.
Approve the risky steps from the dashboard, your phone or Slack.