Ask questions without interrupting a running Codex task

1. Get an answer while the work continues

Codex is changing a component, and you want to know why it chose a particular approach. You can ask that question in a side chat without stopping the main conversation. OpenAI explicitly recommends side chats for progress recaps and explanations during long-running work.

This is useful for understanding an unfamiliar term, checking what a proposed change is meant to achieve, or discussing a tradeoff before deciding whether to intervene. Keep the side question small enough that you can use the answer immediately.

Reference: OpenAI: Long-running work

2. Recognize the temporary Side chat pane

In the earlier Windows package 26.1007.9446.0 trial, the upper-right new-tab control in an active local Codex task opened a temporary pane explicitly labeled Side chat. We sent this guide's copied question there and received an answer while the main task continued. This is an observation of that version, not a promise that every plus button opens Side chat: the new-tab control in the later package opened a browser tab instead.

OpenAI also documents /side and an Open side chat keyboard shortcut. Those routes were not successfully exercised in these trials, so this guide does not present them as tested steps. Use the current documentation to identify your version's control, and confirm that the resulting pane actually says Side chat before sending a question.

Confirm which composer has focus: a message in the main chat can steer the running task. If your version exposes no Side chat action, use a separate saved ChatGPT chat for discussion, then bring back the relevant context explicitly with the tested reference workflow below.

Reference: OpenAI: Slash commands · OpenAI: Commands and keyboard shortcuts

3. Use the inherited context without assuming live synchronization

In the Windows version we inspected, a side chat is created from the current conversation and receives inherited history as reference context. That lets you ask about the task without restating everything from scratch. It is a separate conversation, so the inherited plans are background for your question, not instructions to continue executing the main task.

Treat that inherited history as a starting point, not a live mirror. Do not assume it includes a tool call still in flight or every result produced after the side chat opened. For a progress question, ask what point the answer covers and what it cannot yet verify. Supply the latest relevant result if needed.

Ask for explanation or read-only inspection. A side chat is not a separate Git checkout, and its location in a panel does not make concurrent file edits safe. Leave the main task in charge of the implementation unless you deliberately arrange separate work.

4. Ask a question that helps you make a decision

For an unfamiliar decision, try: 'Explain why this task chose this approach. What problem does it solve, and what is the main tradeoff? Do not change anything.' For progress, ask it to separate confirmed results from work that was merely planned.

If the answer changes your decision, return to the main chat and state the new constraint explicitly. For example: 'Keep the existing public API; apply the internal refactor only.' Do not assume a decision made in the side pane has automatically changed the main task's instructions.

Reference: OpenAI: Long-running work

5. Save useful answers and recognize the chat you opened

Side chats are temporary. In the earlier Windows trial, the pane warned that side chats disappear when the app closes. That warning describes closing the application; it does not establish that simply hiding a pane loses the conversation. Save useful explanations or send the relevant decision back to the main chat before exiting, and follow any close confirmation shown by your version.

A saved ChatGPT conversation and a Mini-created Codex chat are different from that temporary pane. Check the destination and title rather than assuming that all small chat windows share the same history or attachment controls.

6. Bring a saved ChatGPT conversation into the current Codex chat

This route was actually completed in app 26.1013.31256. A test ChatGPT conversation was created, answered and reopened; its reference was then selected and sent in the existing Codex chat. Returning to the Codex chat still showed the sent reference.

In our test, the reference reached Codex with the correct conversation identity, but its read-thread tool returned a placeholder instead of the assistant's reply text. We checked that reply in the original chat's native UI. A successfully attached reference therefore does not guarantee that every message or uploaded file is available to the receiving task: ask it to confirm what it could actually read.

Sending the reference is a follow-up in the main Codex chat and can change its direction. Use a separate discussion first if you only want an explanation; send the reference when you intend to hand that context to the task. The reference does not establish continuous synchronization.

• Use the upper-left mode selector to choose ChatGPT, start a regular Chat and discuss the question there. Note its title; reopening it should show the saved exchange.
• Return to Codex and open the exact task that should receive the discussion. In its main composer, type @ followed by the saved ChatGPT chat's title.
• Click the matching candidate labeled as a ChatGPT conversation. Confirm that it becomes a conversation mention in the composer; plain typed title text is not enough. Our test selected the candidate by an actual click.
• Add a concise request describing what to use from that conversation, then send. Check the sent message's reference and the receiving answer. If the task cannot read the required text or files, provide that material explicitly instead of assuming it was transferred.

7. Use Mini for its own quick conversation

Mini is the floating pet interface. In the tested Windows app, Settings > Mini and pets > Show Mini enabled it. Clicking its compose button opened a small input; sending a bounded question produced a reply and a separate local Codex chat visible in history. The saved chat's history confirmed the exchange.

That test did not turn Mini into the older Quick chat window or attach its local Codex chat to the current task. The tested @ search found the saved ChatGPT conversation, but did not find our Mini-created Codex chat. Use the ChatGPT reference steps above for the attachment workflow verified here.

Reference: OpenAI: App settings · OpenAI: ChatGPT and Codex changelog

8. Treat the older Quick chat launcher as a version difference

OpenAI's projects documentation describes a Quick chat icon beside New chat in the left sidebar and an Add to chat action for bringing a ChatGPT conversation into Codex. That older documented route is separate from Mini and the @ reference workflow above.

The icon was absent and the documented Quick chat shortcut did not open it in our later Windows trial. Inspection of that installed package also found no sidebar launcher or registered command for that route. The updater reported the installed app current, and ordinary ChatGPT chats and @ references worked; no normal setting restored the old launcher.

These observations do not establish an official removal date or rule out account-dependent availability. Do not hunt for a hidden switch, downgrade the app or change account permissions just to follow an old screenshot. Use the tested reference route available in your app, and check the resulting reference before relying on it.

Reference: OpenAI: Projects and chats

Try it in Codex

Understand a decision without changing the task

Use the reference context available in this side chat to explain the main task's current approach.

What problem is it solving, why was this approach chosen, and what is the most important tradeoff? Separate confirmed results from plans. State which point in the task your context covers and any newer progress you cannot verify.

Answer only; do not edit files, run mutating commands, or change the main task. If I need to make a decision, suggest one concise follow-up that I can send in the main chat myself.

Check your result

• You can distinguish temporary Side chat, the older Quick chat entry, Mini and a saved ChatGPT reference.
• For Side chat, the pane identifies itself and the answer states the limits of its inherited context.
• For a ChatGPT handoff, the selected reference appears in the sent message of the intended Codex chat, and you check what the task could read.
• Any implementation decision is explicitly sent back to the main chat; useful temporary answers are saved before app exit.

Sources and scope

Official references support the product behavior described above. The task prompts and practical checks are our community guidance.

• OpenAI: Long-running workSide chats for recaps and explanations without interrupting the main chat; main-chat follow-ups steer work.
• OpenAI: Slash commandsThe desktop /side command starts a temporary side chat.
• OpenAI: Commands and keyboard shortcutsOpen side chat defaults on Windows and macOS; shortcut settings.
• OpenAI: Projects and chatsQuick chat opens an ordinary ChatGPT chat; it is a separate entry point.
• OpenAI: App settingsMini and pet controls; the native messaging result above is a dated Windows observation.
• OpenAI: ChatGPT and Codex changelogThe September 11 Windows Pets quick-chat announcement; it does not establish removal of the older sidebar entry.

Something changed in your version? Send a correction with the guide URL, platform, and version. Leave out credentials and private project content.