Snuck a new "draft" feature into the T3 Code nightly release :) Fixes the "oh shoot I need more info before starting this thread" problem. Surprised how much I've been liking this small change.

Thank you.

Have you been using T3 Code? What are your biggest pain points? Will work on them on my flight home tomorrow :)

so that's why my dev servers looked weird

Why T3 code doesn’t support md files? Or probably any kind of files

I unblocked this with a change to attachments last night. Should be able to do pretty quick now!

Doesn’t work on Windows

> "Should be able to do pretty quick now!" Why doesn't any Windows user know how to read 🙃

bro don’t tell me I have to update AGAIN

We have a nightly release roughly every 3 hours. I just cut another lol I'm sorry

SUNUVA

queue/steer when?

I had a PR for this but Julius wanted to wait for Orchestrator V2 because it's included there

Please add notification or chime when an agent is done working or needs input.

@grimcodes Been thinking about this & was about to open a PR soon This is VERY useful 🙏

Fuck yes I love this

Hey @theo, thinking of building something that lets people run a Claude Code or Codex instance in the cloud, with a sandbox, file system, and a browser for computer-use QA. Been thinking about the UI, and I still think T3 Code is the best control plane for it. Do you see this fitting T3 Code?

I was using t3code right now, and just 5 min ago I was wondering the same thing that why this is not allowed.

Love it, it exposed how many drafts I had 😆

I noticed it too, it's pretty nice

I literally have to update nightly every couple of hours, I swear

THANK YOU

good sign when the small quality of life change is what you're excited about instead of a benchmark number

THANK. YOU.

this already came in handy, what a pleasant surprise feature

OMG, ty 🙏 I literally do this multiple time every day, it was driving me nuts (skill issue on my side tbh)

so much clicking around 😮‍💨

Premature thread creation is a silent killer of context hygiene in agentic editors. The second you launch a run without all the relevant files attached, the model starts making assumptions you end up spending three follow-up prompts correcting. Having a dedicated staging ground to pull in logs and context before committing to a thread is a massive UX upgrade for agent workflows. Does the draft state persist across app restarts if you leave a prompt sitting overnight?

Why not build that into that stash feature in the composer? Creeping complexity in the sidebar can make it overwhelming to get a overview.

Bro i’m gonna have a fuck ton of drafts now even more so than the hundreds of issues/epics/milestones i already plan with lol

Ngl, I was quite critical of T3 Code before, and now I went and tried again and it's blowing me away. UX/UI also feels alot better. Well done guys!

Nice, been finding this confusing in Codex and sometimes lost part of “draft”, especially pasted images. Would be awesome to be able to que up drafts on the phone and then assign to device/model etc later. Inbox of prompts to shoot off later. Now sitting in scattered notes.

I encountered an issue when calling Codex CLI using the T3 code. Will this cause any account anomalies? Also, when will the consumption analysis feature be available for the Windows version? Haha, thanks for the hard work.😃

This is great! I was using the fact that the prompt is saved even if you switch threads to accomplish the same thing, glad I can actually see them in the sidebar now! I lost a task at least 3 or 4 times just to discover it in the new thread window on that project after a week 😃. Can you add a feature to add comments to any part of the response? I have been keeping a forked version of t3code just cause I added a vibe coded version of this feature. I always like to add comments with context to the agent’s answer.

God bless. This is gonna help me so much....

draft sounds small, but it saves you from keeping half a prompt in your head while gathering context

This is random. But i am working on hot loading context from previous sessions in github copilot studio. Did t3 solve this somehow. Trying not to reinvent the wheel.

What do you mean by “hot loading context”?

Basically need to be able to resolve previous separate sessions and dynamically load context from those sessions without using generic harness memory. Its kind of a weird question but business people do not understand the concept of resuming a session and will start a completely new session despite the model having over 500k tokens of context left before it overflows. So I called it hotloading context into the session. Currently using onedrive for this. But curious how you solved it for non technical users.

This is really cool. Quick Q though - what’s up with that hover state being 5 miles higher than the pointer? Recording bug or styling?

@theo can you add pi support already please mate

How do we get nightly build?

Theo, could you please clarify whether I can use my Claude Code subscription with T3 Code through my Claude subscription? I’m a bit confused about all these restrictions/blocking issues from Anthropic. I’d really appreciate it if you could clarify how this works.