Saeta is in development and has no public release yet. This page describes the current development build and our plans.

Agent decides

Agent decides lets an agent's own permission system judge its calls, while Saeta keeps its blocks and its own protections. When the agent would ask you something, the question comes to Saeta's inbox, so you can still answer from the desktop or your phone.

It works for Claude Code and Codex, whose own questions reach Saeta through a hook (Claude Code's permission prompt, Codex leaving its sandbox). Other agents' questions don't, so Saeta refuses it for them.

What decides

  1. Saeta's blocks and its own protections, and an organisation's rules, as in how Saeta decides.
  2. Your rules, then Saeta's read-only allows.
  3. Everything else goes to the agent and its own permission settings.

When nobody answers one of the agent's own questions in time, it goes back to the agent, which asks in its own prompt.

Turn it on

In the desktop app, use the Agent decides column on the Agents screen. From a terminal:

saeta agent-decides on claude-code
saeta agent-decides off claude-code
saeta agent-decides               # show it

It applies to every session of that agent, everywhere. It also takes over from safe auto mode for that agent: in a project in safe auto mode, that agent's calls go to its own permission system, not to safe auto mode's list.