How it works
Alter Ego is a connector, so in these tools the model itself decides when to call it. In claude.ai and in the regular ChatGPT chat, loading your memory with start and adding a journal line with log happen by themselves. In coding agents and in Work mode the model is focused on the task at hand. For a short question or a quick code change it may go straight to the answer and leave out start and log.
Name Alter Ego at the beginning of the chat and say what you expect. That gives the model a clear instruction to load your memory and keep your journal. If it still skips a step, a short reminder in the chat is enough.
The opening line
Begin the chat with one short line that mentions Alter Ego and says what you expect. You can copy this one.
Use Alter Ego for this chat: start, and log after every answer.Add your task straight after it, in the same message. To work on a client or project, name it: Use Alter Ego for this chat, open the context Acme, and log after every answer.
Per tool
ChatGPT Codex
Type @ and pick the Alter Ego chip in the message box. The chip tells Codex that this chat uses the connector. Then write your opening line and your task.
ChatGPT Work mode
Mention Alter Ego with the @Alter Ego chip in your first message, followed by the opening line.
Claude Code
Add the connector once with your Alter Ego connector URL, then start a session with the opening line. Claude Code lists connected servers with /mcp, so you can check that Alter Ego is available. The docs cover adding any remote MCP server.
Per chat and optional
You do not need to put this in account settings, custom instructions or project files. The opening line works per chat, so you choose for each conversation whether your memory is part of it. Your rules and profile still load from Alter Ego itself once the chat starts.
Quick questions without the line are fine too. Use it whenever a chat should read your memory and write to your journal.
More
- Get started for adding the connector in Claude, ChatGPT and Gemini.
- Tools for what
start,open_contextandlogdo. - Troubleshooting if the assistant does not use Alter Ego.
- Teach it once for saving rules, contexts and playbooks.