You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Automatically spin up a portless server with the url of the git worktree, so {worktree}.{project}:1355 and you could just add an action to open this in your default browser quickly from t3code
This has worked well for me in local development
Reacted by Max Liefkes, oliver, Erik Slovák, Robin Lamberti, Rodrigo Rocha and Saikrishna Ambeti
I built this for myself, I’m sharing it via PR (#963) in case any part of it is useful as a starting point or source of ideas. The PR also includes a reconstruction plan and screenshots.
The current browser runtime keeps only the 3 most recently used web views alive and evicts the rest. That tradeoff was mainly to keep memory usage bounded, and it could be made smarter.
This looks a lot like cmux in the video from @acoiro. To be honest I like it if it stays lightweight.
If t3code is enabled to be integrated into cmux as a first citizen app and cmux integrates the project->chat/task based side bar part it can be combined into each other and it would "solve" the browser question for everybody who wants it. And it includes a little more...
I built a repo that forks both projects and adjusts them as little as I could to integrate them into each other. for t3code this means primarily changes on how a chat can be started without the sidebar being active.
Yes. Will put the browser thing on the roadmap in the short-ish future.
Just need to figure out auth cause thats where most of the current solutions breaks down that i can't sign in to any service which makes them pretty unusable for any real work
Reacted by nick, Ben Macpherson, Vincenzo Rosciano, FeliΞ, Saikrishna Ambeti and Laurynas Gel
@juliusmarminge totally get the auth concern - that's the tricky part for sure.
On the competitive side though, Cursor keeps shipping hard on this front. Their built-in browser is getting better and better, and they're even planning to add an iOS emulator directly into Cursor. We're falling behind here and I think this needs to move up in priority.
I'd love to help push this forward. Two options from my side:
You set the direction - share your vision for how the browser should work (architecture, auth approach, UX) and I'll pick it up and start implementing based on that.
I prepare a plan first - I can put together an implementation plan / proposal and share it with you for review before writing any code.
Whatever works best for you.
Also, one thing I've been thinking about beyond just the browser - a custom Vite plugin that would help the design tool understand which component is currently highlighted, map DOM elements back to source components, etc. Kind of like what Cursor's design mode does. But obviously the browser itself is the prerequisite and needs to land first before any of that makes sense.
yea so for design mode we could use some of Aidan's work in react-grab or whatever right? tanstack devtools also injects source paths to components.
for the browser view, I wanna improve the layout system first (#2011) so that you're not limited to where you place the browser. if you wanna work on the actual brwoser implementation in the meantime be my guest :)
@juliusmarmingePR #963 from @acoiro was closed without a code review. Since the browser implementation on the roadmap, could you take a quick look at that PR and share what you'd want done differently? Would help us avoid going down the wrong path.
Activity
nckre commentedon Mar 7, 2026
@juliusmarminge would recommend checking out portless from @ctate as a light alternative.
Automatically spin up a portless server with the url of the git worktree, so {worktree}.{project}:1355 and you could just add an action to open this in your default browser quickly from t3code
This has worked well for me in local development
acoiro commentedon Mar 12, 2026
I built this for myself, I’m sharing it via PR (#963) in case any part of it is useful as a starting point or source of ideas. The PR also includes a reconstruction plan and screenshots.
The current browser runtime keeps only the 3 most recently used web views alive and evicts the rest. That tradeoff was mainly to keep memory usage bounded, and it could be made smarter.
in-app-browser.mp4
TimoKruth commentedon Mar 27, 2026
This looks a lot like cmux in the video from @acoiro. To be honest I like it if it stays lightweight.
If t3code is enabled to be integrated into cmux as a first citizen app and cmux integrates the project->chat/task based side bar part it can be combined into each other and it would "solve" the browser question for everybody who wants it. And it includes a little more...
I built a repo that forks both projects and adjusts them as little as I could to integrate them into each other. for t3code this means primarily changes on how a chat can be started without the sidebar being active.
Project link: https://github.com/TimoKruth/cmux-t3code
akarabach commentedon Apr 9, 2026
@t3dotgg @juliusmarminge @maria-rcks design mode in Cursor 3 is amazing
juliusmarminge commentedon Apr 9, 2026
Yes. Will put the browser thing on the roadmap in the short-ish future.
Just need to figure out auth cause thats where most of the current solutions breaks down that i can't sign in to any service which makes them pretty unusable for any real work
akarabach commentedon Apr 14, 2026
@juliusmarminge totally get the auth concern - that's the tricky part for sure.
On the competitive side though, Cursor keeps shipping hard on this front. Their built-in browser is getting better and better, and they're even planning to add an iOS emulator directly into Cursor. We're falling behind here and I think this needs to move up in priority.
I'd love to help push this forward. Two options from my side:
Whatever works best for you.
Also, one thing I've been thinking about beyond just the browser - a custom Vite plugin that would help the design tool understand which component is currently highlighted, map DOM elements back to source components, etc. Kind of like what Cursor's design mode does. But obviously the browser itself is the prerequisite and needs to land first before any of that makes sense.
juliusmarminge commentedon Apr 14, 2026
yea so for design mode we could use some of Aidan's work in
react-grabor whatever right? tanstack devtools also injects source paths to components.for the browser view, I wanna improve the layout system first (#2011) so that you're not limited to where you place the browser. if you wanna work on the actual brwoser implementation in the meantime be my guest :)
akarabach commentedon Apr 15, 2026
@juliusmarminge PR #963 from @acoiro was closed without a code review. Since the browser implementation on the roadmap, could you take a quick look at that PR and share what you'd want done differently? Would help us avoid going down the wrong path.
refactor(web): host the sandbox control in the right panel (pingdotgg#37
chore: sync with upstream pingdotgg/t3code main (pingdotgg#37)
Merge pull request #37 from i2cjak/backplane/cross-machine-rooms
Merge pull request pingdotgg#37 from Yarden-zamir/fix/server-trusted-…
ci(fork): run fork tests in Fork Sync and report Fork CI failures
chore(fork): untrack local links from _shared
Merge pull request 'fix(server): T3CODE-29 outbound connections survi…
Merge pull request 'fix(web): Grow's shortcut keys no longer die when…