AI Error WikiSearch

Codex: stream disconnected before completion

The connection to an in-progress Codex task ends early.

Quick answer

The response stream ended before Codex reported a normal completion. In a Windows desktop and VS Code report, the user saw repeated reconnect attempts followed by this error. The message does not tell you which local edits, if any, were already written.

Exact message

stream disconnected before completion: An error occurred while processing your request.

What it means

The chat and the working directory can be out of step after a disconnect. Treat the task as unfinished until you inspect the last visible action and the files it touched. The cited report says changing clients helped only briefly for one person, so a client switch is not a proven repair.

What to try

  1. Read the last completed action in the task and inspect the affected files. Save any useful output before retrying.
  2. Ask Codex to continue from the current state or retry only the unfinished part. If the previous request could have sent data or changed a remote service, check that service first.
  3. If disconnections repeat, note the client, OS, time, model, and any request ID. Restart the client once if practical, then record whether the behavior changes.

User report

The issue describes Windows Codex Desktop and VS Code disconnecting after several Reconnecting attempts. A temporary improvement in the VS Code extension did not last for that reporter. Their technical log counts describe events in one local setup, not how often the error happens for everyone.

FAQ

Did the interrupted task finish?

You cannot tell from the stream message. Open the output file, check the changed lines, or inspect the task’s last action. A task may have written some files before the connection dropped. If it was supposed to create a downloadable artifact, check whether the artifact exists before asking for another copy.

Can I resume the same run?

If the client offers Continue or Retry, use it after checking the current files. Otherwise, start with a short follow-up that says what is already done and what remains. Avoid asking it to repeat the entire task blindly; the first run may have made partial changes even though the reply never finished.

Is this the same as model capacity?

No. A capacity notice appears when the selected model cannot take a request. A stream disconnect describes a response that was already in progress and did not finish normally. The practical difference is important: after a disconnect, inspect possible partial work before resending anything. See the capacity guide if that is the exact message on your screen.

Sources