One gate between your agent and the world
Here is what happens when an agent wants to send an email.
- The agent asks first. Before a risky action, it tells Gatesi the tool, the action and the details.
- Gatesi checks your rules. Each request is allowed, sent to you for approval, or blocked. A blocked request is refused at once, and the agent is told why.
- You get a message on your phone. Telegram shows the request in plain language, with its risk level, the exact details, and Approve and Deny buttons.
- You decide. Approve once, deny, or allow for an hour. If you don't answer, Gatesi sends a reminder, then refuses when the wait ends. Your biggest actions can need a second approver.
- One approval, one action. An approval covers only the exact request you saw. If the details change, or it was already used, it is refused.
- Everything is recorded. Requests, decisions and results go into a hash-chained trace that you can replay, verify, or sign as a receipt.
Two ways to connect
Quick start. The agent asks Gatesi before it acts. You are up and running in a few lines.
With the vault. Gatesi makes the call and adds the stored key, so the agent never sees the credential. Use this for the actions that matter most.
Gatesi checks every action your agent sends through it, and with the vault the agent never holds the keys to those services. Combine rules, block lists and the vault to set hard limits on what an agent can do.
An optional AI note adds a one-line plain-language summary to each request, so you can decide faster.