A kill switch
for your AI agents.
When an agent goes sideways, you should not have to race to the terminal. Flip the kill switch from your phone and every gated tool call is denied until you say otherwise.
Mis en avant sur Hacker NewsOne move takes it all back
However autonomous your agents are, one person should be able to halt any of them immediately, without negotiating with the agent first.
One tap to halt
Flip the kill switch from your phone. You do not have to find the terminal, kill the process, or argue with the agent first.
Denies everything gated
While it is on, every gated tool call is denied, including the ones your policy would normally auto-approve. Nothing slips through.
From wherever you are
The switch lives on your phone, so a runaway agent does not depend on you being at your desk to stop it.
Stays off until you release it
The session stays halted until you turn the switch back off, so a stuck agent cannot quietly resume on its own.
Works across your agents
Halt a specific session whether the work is running in Claude Code, Codex, Cursor, or Hermes.
Why you can grant autonomy
The kill switch is not an admission the system failed. It is the thing that lets you hand an agent real freedom, because you can take it back in one move.
Keep the cord in reach.
Connect an agent and the kill switch is one tap away on your phone.
Start your trialGuides
More control
Frequently asked questions
What is an AI agent kill switch?
It is a single control that immediately stops a running AI agent. With Pushary you flip it from your phone, and the agent's gated tool calls are denied until you release it.
How do I stop a runaway agent?
Open Pushary on your phone and turn on the kill switch for that session. From that moment every gated tool call is denied, so the agent cannot run any command that goes through the gate.
Does it stop actions I would normally auto-approve?
Yes. That is what makes it a kill switch and not just a stricter policy. While it is on, even the commands your policy auto-approves are denied, until you turn it back off.
Does the agent resume on its own?
No. The session stays halted until you release the switch yourself, so a stuck or looping agent does not quietly pick back up.
Which agents does it work with?
It works with the agents Pushary can gate through hooks, including Claude Code, Codex, the Cursor plugin, and Hermes. Plain MCP clients still get notifications and questions but cannot be hard-stopped the same way.
How much does it cost?
The kill switch is on the Agent Pro and Team plans. Paid agent plans start at $9.99 per month with a 3-day trial.
Shipping an AI agent to your own customers? Rules can ask your customer instead of you, for example before a refund of $500 or more. See the Partner plan or the human-in-the-loop API.