Connect an agent

What happens next: the agent appears as @claude (or similar) and remembers in a bullet under Home called Claude’s memory, unless you unticked that on the consent page. Beyond that it reads your whole outline, and changes it only if you approved read and write. You can see every agent, move its memory to any bullet you are in, and take back every key under Agents and API keys in the app. For a narrower reach, hold the agent to one bullet with one press: open that bullet, open Agents and API keys in Settings, tap the agent’s row under Your agents, and press Let it reach only “…”, the bullet you are in. It then reads and writes that bullet and everything under it, and remembers there, even if you approved read only. It reaches nothing else in your outline, and its row says so. An agent you create there with “Let it read and write this outline” and “Give it a bullet of its own to remember in” both unticked reaches nothing until you give it a bullet the same way. Its key goes in a Bearer header, which clients such as Claude Code, Codex or Cursor send. ChatGPT cannot send a key. The Claude apps can send one only through the Request headers setting of a custom connector, which Anthropic calls a beta for a limited set of organizations; an administrator enters the key there, and that one key then serves every member of the organization.

A good first job for it: the inbox pattern. The agent works, and writes anything only you can do into your tree as checkboxes with the why in the note; you tick, it continues. The template gallery has it ready as “Inbox (from your agent)”, prompt included, and the story of why is a two-minute read.