Coming soon

Approve what your AI coding agent does, before it happens.

Saeta is one approval inbox for local coding agents: Claude Code, Cursor and others. Answer on your desktop, or with one tap on your phone. No more babysitting a terminal, no more --dangerously-skip-permissions.

We'll only email you about the Saeta launch. Unsubscribe anytime. Privacy notice

How it works

  1. Intercept

    Claude Code, Cursor and Codex use hooks, and OpenCode and Pi use Saeta's plugins, so Saeta can pause their shell commands, file edits and tool calls before they run. Gemini CLI connects through MCP: asking Saeta depends on the agent, so this connection is not mandatory interception.

  2. Decide

    Known-safe actions pass instantly against your trust list. Anything else pauses and lands in one inbox, on your desktop or on your phone if you've stepped away.

  3. Continue

    Approve, deny, or reply with a correction such as "use pnpm instead", and the agent picks up exactly where it left off.

Safe auto mode

Let the agent work. Saeta still stops the risky calls.

Turn on safe auto mode for a project, or for one agent session, and everyday work runs without asking you: builds, tests, installs and edits inside the project.

Saeta still asks before a call that would:

  • delete files
  • throw away work in git
  • push to a remote or publish a package
  • connect to another machine or upload data
  • run as root or change file permissions
  • read or edit files outside the project
  • read secrets such as .env files and tokens
  • run a command hidden inside another

If you don't answer in time, the answer is no. Saeta's blocks, such as reading private keys or rm -rf /, still block. Saeta reads each command line, not what a script does once it runs, so a build or test script can still do anything you could.

Or let the agent decide

For Claude Code or Codex, you can turn on Agent decides instead. Saeta's blocks, the protections for Saeta itself and your own rules still apply; everything else is left to the agent's own permission settings, so it is only as careful as they are. When the agent would ask you, in Claude Code's permission dialog or when Codex wants to run a command outside its sandbox, the question comes to Saeta's inbox and your phone instead. No rule answers it, and the answer is never remembered. If you don't answer in time, the agent asks in its own prompt.

Built for how you actually work

End-to-end encrypted

Your code, file paths and commands never reach our servers in readable form. The cloud only relays encrypted data.

Local-first

The desktop app keeps working offline. Approvals, the trust list and the activity log need no network.

Same guardrails, any agent

Your trust list lives in Saeta, not in the agent, so the same guardrails apply when you switch agents or run several at once. Claude Code, Cursor and Codex through their hooks, OpenCode and Pi through plugins, and Gemini CLI through MCP.

A real free tier

The inbox, agent wiring, trust list and activity log need no account and no card. Ever.

More than yes or no

Approve or deny, pick one of several options, or type a reply the agent follows.

Every decision on record

The activity log shows what each agent asked, what you answered and which rule matched. It stays on your computer.

A guardrail, not a sandbox

Saeta catches the mistakes agents make: a hallucinated rm -rf, a misread instruction. It is not a security boundary against a hostile agent, and we won't pretend it is.

What Saeta is and isn't

Questions

What is Saeta?

Saeta is a desktop app for macOS and Linux that pauses your AI coding agent before risky actions and asks you to approve or deny them, in one inbox on your computer or with a tap on your phone. Windows support comes later.

Which AI coding agents does Saeta work with?

Claude Code, Cursor and Codex through their hooks, and OpenCode and Pi through Saeta's plugins: the call waits for your answer. Gemini CLI connects through MCP, so it asks Saeta only when the model chooses to. Your trust list lives in Saeta, not in the agent, so the same guardrails apply when you switch agents or run several at once.

Can Saeta read my code?

No. Approval requests are end-to-end encrypted between your computer and your phone. Our servers only pass along encrypted data they cannot read.

Is Saeta a sandbox?

No. It is a guardrail against agent mistakes, not a security boundary against a hostile agent. Read what Saeta is and isn't.

Is --dangerously-skip-permissions safe?

Only in an isolated environment, such as a disposable container. On your own machine it lets the agent run any command or edit without asking. Saeta lets you keep permissions on without babysitting the terminal.

How is Saeta different from Claude Code's own permission prompts?

Claude Code asks in the terminal where each session runs. Saeta puts the requests from every agent in one inbox on your desktop and phone, and lets known-safe actions through automatically with your trust list.

When does Saeta launch?

Saeta is in development. The phone app launches on Android first, with iOS soon after. Join the waitlist and we'll email you when early access opens.

Be first to know when Saeta ships.

We'll only email you about the Saeta launch. Unsubscribe anytime. Privacy notice