The Composer predictions feature suggests a next message based on the conversation in your current Codex desktop thread. It can help you continue a task, refine a draft, or ask for a useful next step after Codex responds. Review each suggestion to decide whether it matches what you want to do.
Before you start, update the Codex desktop app and sign in to an eligible account. Check Composer predictions in Codex for current plan, model, and thread requirements.
Tab to accept
- Send a message in a supported thread. When Codex finishes responding, look for a suggestion in the message box.
- Press Tab to accept the prediction. The suggested text becomes an editable draft in the message box.
- Check the suggested action, audience, and details. Edit anything that doesn’t match your intent, then send it when you’re ready.
To write something else, start typing your own message. You can keep typing without waiting for a prediction. Accepting a prediction adds text to the message box. It does not send the message for you.
A prediction won’t appear after every response. You can continue the conversation by typing your own message.
Example: Refine a project update
In this example, you’ve asked Codex to summarize project notes. A prediction suggests turning the summary into a team update. Accept it if that’s your next step, or edit it to specify the audience and what to include.
For example, you could edit the accepted suggestion before sending:
Turn this summary into a short update for the project team. Separate confirmed progress from open questions, and don’t describe the launch date as final.
Review the resulting update against your notes before sharing it. Accepting a useful suggestion doesn’t verify the facts in the response.
How predictions use context
Predictions use the conversation in your current thread. If context from past work or connected apps is already in that thread, it can inform a prediction. Predictions don’t look up memories or retrieve information from connected apps on their own. If you have memory enabled, memory context already in your thread can personalize predictions. You can use predictions with memory off.
Availability and controls
For current availability, how to turn predictions on or off, usage limits, billing, and troubleshooting, see Composer predictions in Codex in the Help Center.