[Feature]: Add a Codex-style /side command

[Lermatroid]

It would be great to have something like Codex’s /side for quick ephemeral side thread from the current conversation to explore a question without derailing the main thread in the side panel.

Example of what this looks like: https://x.com/Dimillian/status/2049929842133520577

[lastobelus]

This is particularly useful when your side query

a) needs the main thread context and
b) will have verbose output, and therefore trigger a compaction (or add significant token cost to subsequent turns) if you ask it in the main thread

[joshua-redmond-1st]

Yes! I really miss the side chat/btw feature when I use T3 Code

I love T3 Code, but the rest of my team doesn't want to use it because it's not got this feature lol

[jackelly7]

Yes, i completely agree. I would love a codex app inspired side chat mode. The codex app ui for it is great, as well as the keyboard shortcuts. Codex: /side CC: /btw

[iAmAdheil]

Really miss /btw when using t3

[Luzivog]

I'd like this too. My use case is asking a quick question about the current task while the main agent continues working, without steering or interrupting that task or having to re-explain the context in a new top-level chat.

The useful behavior would be:

• Open a side chat from the current conversation, with the relevant context available.
• Ask follow-up questions in a separate side panel while the main run continues.
• Keep the side conversation separate from the main thread, with a way to bring an answer back when useful.
• Make it easy to open and dismiss through a button or keyboard shortcut.

For example, while the agent implements a change, I want to ask why it chose an approach or explore an alternative without turning that question into new instructions for the implementation. Codex-style side chats are the interaction I have in mind.

[nbros]

Another Codex feature I use constantly in Codex is “More Details” on selected text.
I think essentially this is just a prompt to "explain this selection" with the original thread as context.
Currently in t3 code I haven't found a way to do this.
Either I pollute the original thread context with my question, or I create a separate thread for my question, but then I have to copy the original context manually, since t3 code also seems to be lacking a "fork thread" feature.
I would love a quick and easy select → explain → popover interaction in T3 Code like in Codex.

[hardik88t]

Having /btw for quick question like in claude code and /side for a seperate window to chat with the current running agent context like codex would be great. logic for both would be similar just different UX. both are great.
maybe I will try doing something in t3code for myself. will share if i make something useful.

[LukasFB]

+1 to this.
One more reason to add this: It's also very token-efficient because it hits the main thread's cache. Asking the same question in a new thread not only requires rebuilding the context (slow) but is also a complete cache miss (expensive).

[bribeck]

+1, on desktop and mobile. I'd love to ask a quick side question after any response that sees the full thread, but disappears afterwards and never enters the main thread's later context (exactly what #9274 described).

[rohan-patnaik]

Continuing here as requested after #12794 was closed. I use SideChat to ask a quick question from a snapshot of the current conversation while the main agent keeps working, with independent tabs/models and an explicit send-back action. The draft has before/after screenshots showing that workflow. Happy to align it with the V2 rewrite; keeping the proposal here for now.

[dantheuber]

+1 from me as well. It is very useful to be able to interrogate the main thread and have q & a using cheaper models in a side chat that doesnt add anything to the main threads context. Been using this a lot in the codex (ChatGPT) desktop app, would like to be able to do the same in T3 code

[kush-daga]

+1 to this, it has been really helpful. Apart from just side-chat, one more very helpful feature that just codex has is the "More Details" option that comes up when you highlight something. Would be a great addition here!

[rodonguyen]

What is the concensus to green flag this feature guys? We have good amount of Upvotes here.

[wojtekpiskorz]

yeap /side /btw is a must!

[nbros]

Yes please 🙏

[zzz-temp]

+1

[jeetganatra]

+1 to having /btw As models become more and more long running and autonomous and don't provide much mid turn messages to the user (such as Fable 5.1 and Opus 5.5) this feature can be really helpful to know what is happening and if the work has derailed in anyway

[Lightning11wins]

I'm looking into using t3.code as my control surface instead of calling Claude Code directly and the lack of /btw is one of the biggest things holding me back. I love just being able to ask random questions about how the work is going, what was that permission error, etc. without distracting or interrupting the main model. This feature would be super helpful!

[tarushvkodes]

Yeah I too find myself wishing I was able to do something like /side in codex. Would love to see this feature

[sankalpmukim]

+1 — would really like this.

[Lizzergas]

+1

Although you can fork the chat and use it as the /side, but it is convoluted

[andrewslabbert]

I use T3 Code with Codex and would really value this. While the agent is working, I want to ask questions about its approach without interrupting it or adding the side discussion to the main context. Forking works, but a lightweight side panel would make this much easier.

[soumen02]

+1 — this is one of my favorite Codex features, and I would really like it in T3.

While an agent is working on a task, I want a parallel conversation with that task's context for curiosity, learning, checking assumptions, or understanding a decision. For example: “Why did you choose this approach?”, “Explain the code you just changed,” or “What risks should I check?” I want to explore follow-up questions without interrupting the implementation or making that detour part of the main thread's future context.

A small first version would already be useful:

• Open a side chat from the active thread with a button or keyboard shortcut (/side would also be welcome).
• Inherit a snapshot of the parent conversation and show where that snapshot was taken. If forking while a run is active requires using the latest stable boundary, that is fine; make the boundary clear.
• Let the main run continue independently. Opening, using, or closing the side chat must not steer, queue messages into, cancel, or restart the parent run.
• Support a real multi-turn side conversation, so I can keep asking follow-up questions.
• Keep side-chat messages and answers out of the parent's future context unless I explicitly send something back.
• Show it beside the task and make it easy to dismiss, without cluttering the normal thread list. An explicit “Keep as thread” action would be useful.

For scope, one side panel using the current model would be enough initially. Multiple tabs, a separate model picker, and richer merge-back controls could come later. An existing V2 fork/context-transfer path seems worth evaluating as the foundation, if it supports these semantics; there is no need to revive the pre-V2 implementation wholesale.

A useful acceptance scenario: start a long implementation task → open a side chat while it runs → ask an explanation and two follow-ups → close the side chat → continue the main conversation. The implementation should have kept running throughout, and the side exchange should not have entered the main conversation's model context. Sending back a chosen answer should require a separate, explicit action.

One detail worth making clear in the UX: an isolated conversation does not necessarily mean an isolated filesystem. For this learning/checking workflow, a side chat that starts with read-only tools (or no tools) would be useful, so it does not accidentally compete with the main agent's edits. The context snapshot is also a point-in-time view, while workspace files may continue changing.

The core value for me is being able to understand and interrogate ongoing work without derailing it. I would use even that minimal version frequently.

[antoniel]

I opened #26454 to track this request.

If anyone wants to spin up Claude to draft a spec in the issue comments, that would be great. It could help turn the ideas from this discussion into a concrete proposal.

[t3dotgg]

/btw

[atticoos]

THANK YOU!

[rodonguyen]

Nice!