Guard any agent in one HTTP call.
Before your agent runs a tool, ask the Guard API: allow, ask a human, or deny. Works from any language and any framework. Free.
Get a free agent key
I already have a key
The key never leaves this browser except to agentleisure.com. Checks without approval need no key at all.
Would this tool call be allowed?
What your agent asked for
Add a key to see the approval requests made with it.
Only the human with the approval link can approve or deny. This key can read status, never decide.
Free or Team
Add a key to see your plan. Free: the newest 100 journal entries and 1,200 checks a minute. Team, 29 USDC a month: the full journal history and export, 6,000 checks a minute, priority support.
Every decision, signed and chained
Add a key: every Guard API check made with it is recorded here, with a hash chain and a server signature.
Approve and Deny from Telegram
When your agent asks for approval, the request arrives in Telegram with buttons. One tap decides it.
Add a key to connect a chat.
Is this page, issue or comment trying to steer your agent?
Paste anything your agent is about to read: a web page, a GitHub issue, a PR comment, an email, an MCP tool description. Nothing is stored.
Copy, paste, done
curl -s https://agentleisure.com/api/guard/check \
-H "Authorization: Bearer YOUR_AGENT_KEY" \
-H "Content-Type: application/json" \
-d '{"tool":"run_sql","args":{"query":"DROP TABLE users"},"request_approval":true}'More: agent-guard for Node · Claude Code hook · Python client · OpenAPI