Version 0.4 is HERE!

[zemaj]

Apologies for the gap in releases - we’ve been cooking, and we hope you like what we've created 😋 This update touches almost every part of Code. There’s virtually no area that hasn’t been refined or rebuilt, with major refactors under the hood to make everything smoother, faster, and smarter.

We’ve rolled everything from the planned v0.3 straight into v0.4 - it’s that big. Here are some of the highlights:

---

Massive Upgrades to Auto Drive

We’ve been working really hard on this one, and hopefully the results speak for themselves. Auto Drive is now significantly smarter at writing code, orchestrating agents, and running checks on itself.

It’s especially powerful for large updates - my favorite workflow is to plan first, then simply use /auto and let Auto Drive take over. Leave it running overnight, and you’ll usually wake up to a finished project.

We’re using Auto Drive in about 60% of our sessions now, and it only gets better with more agents connected. Seriously - give it a try.

All Settings, One Command

As Code grows more capable, we’ve made configuration simpler. Everything now lives under the unified /settings command, so you can easily see what’s enabled and toggle features on or off as you work.

New UX Cards

With Auto Drive and multi-agent workflows ramping up, visibility became key. We’ve introduced a new card-based UI for Agents, Browser, Search, and Auto Drive, grouping related actions into compact, elegant views - and expanding into overlays when you need deep detail. It’s clarity and control, side by side.

Huge Performance Gains

We’ve hunted down and fixed CPU and memory bottlenecks that could slow things down when running many agents.
Now, Code stays buttery smooth, no matter how ambitious your session gets.

Smarter Agent Targeting

You can now target specific models directly via CLI commands like /plan or /code. Use gemini-flash-2.5 for lightning-fast codebase research, or claude-sonnet-4.5 for complex problem solving. This will soon expand to support any model or CLI from the settings page.

---

And that’s just the start. There’s a lot more on the way.
We’d love to hear what you want us to build next - so drop your thoughts, ideas, and wildest feature requests.

[GraceKellySS]

Is there an auto drive documentation/tutorial/demonstration?

[zemaj]

Putting together a video now :)

[yinkev]

Putting together a video now :)

Thank you :)

[zemaj]

[Edited] That last video had some encoding issues, so uploaded a new one https://youtu.be/UOASHZPruQk

[GraceKellySS]

Here you go https://youtu.be/UOASHZPruQk

Thank you!!

[elsbrock]

Fantastic release, thank you!

I noticed that you can do /new during a running conversation, is this new too? It's a little confusing how this works, but seems indeed to run parallel conversations and no way to switch between them.

[zemaj]

Ah that would be a bug. It's just the existing turn completing in the background still outputting messages. Will fix!

[zemaj]

I left Auto Drive running overnight working on the outstanding issues for Code and it triaged all 30 of them and solved the 6 most important ones. Have not found any issues with the solutions so far. It solved one issue which was "by design" and not a real issue, so will drop that, but everything else is perfect. Got to 30 turns (6hr) before it seems like it hit up against a memory issue. It looks like it used 32 agents. Agents are super effective at both reducing token usage (surprisingly) and performing high quality work, particularly when it runs them in parallel. It looks like large numbers of agents are still causing memory leaks so will look into that.

I've had it running much longer in the past, but that was pre-agents and pre-diagnostics. It tended to go off on tangents and was more hit and miss. It's getting far, far better at sticking to the goal and doing real productive work. I rarely have a "bad" session now.

Going forward it's going to be important to make the work done more visible. With it running so long we need better oversight. So next step will be to add an interface to make it clearer what it's actually done so you can go through and review it properly.

[zemaj]

In comparison, previously I would have had to create a separate branch for each, push code through on each of them for several hours and it would easily have taken up a whole day to do this work.

[zemaj]

This is so cool. I left it working all day on outstanding issues. It discovered by itself that the codex code review would run and is pinging it and working on it's PRs and waiting for results #358

Prompt was literally just "work through open issues at https://github.com/just-every/code/issues using gh". At 42 turns and submitted 7 PRs now! Merged a few already and will monitor the remainder.