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
{{ message }}
Repository navigation
Setting to prevent TUI scrolling when new message are streamed-in #7648
Title is self-explanatory.
I like to read TUI messages while the Agent is working and streaming-in new messages below, but the current behavior of scrolling down is a bit annoying.
I searched for and didn't find any setting that controls that.
Reacted by Karolis Šarapnickis, Adam Spiers, Tilman Bregler, Anton Ukhanev, javioverflow, Wei, Danny, Radu Salagean, ecavallin, Eugene Paniot and 17 more
Ideally, this feature would be intelligent enough to disable auto-scroll only when the user has intentionally scrolled up from the bottom of the window. This is how many other LLM chat interfaces work. That way, if the current scroll position is at the bottom of the stream, then the auto-scroll will stay active because it's safe to assume the user wants to see the real-time stream. But if the user has moved the scroll position to a previous point in the stream, then the assumption should be that the user wants auto-scroll to be disabled.
Reacted by Nicholas Grisafi, Adam Spiers, Anton Ukhanev, Daemon, Danny, marcus-stravito, write2art, piknockyou, Karel Peeters, Felipe Alves and 1 more
This issue does not seem to happen when using Kimi K2.5, but it happens constantly with Claude. So it appears this is caused by how OC is interpreting and rendering Claude's (and mabye other models) specific responses. I notice that Kimi K2.5 streams its responses a bit differently; so it should be easy to visually see the difference.
yes, I agree - this automatic scrolling down when the AI response changes makes this tool a little frustrating to use - I'm being forced to wait for some lengthy responses (so often a lot more than I wanted) until the Agent is completely finished. Then I can begin reading the response. It makes the process so much slower :-(
Otherwise - TOTALLY LOVING this tool - really really awesome work guys
Tested with MiniMax-M2.5 on macOS in iTerm2 — does exactly what's asked for in this issue: scroll up (wheel or PageUp) pauses sticky, scroll back to the bottom resumes it.
Edit: On closer testing I can't actually reproduce the bug without the file either — mouse defaults to true and the desired behavior already works. So this might not be the actual fix. Sorry for the noise.
Reacted by Lauri Nõmme, Rong Zhou and Robin Bially
Activity
github-actions commentedon Jan 10, 2026
This issue might be a duplicate of existing issues. Please check:
Feel free to ignore if none of these address your specific case.
alexx-ftw commentedon Jan 10, 2026
I have implemented this feature in PR #7651. You can now disable auto-scrolling by setting
tui.auto_scrolltofalsein your config.JuliaBonita commentedon Jan 11, 2026
Ideally, this feature would be intelligent enough to disable auto-scroll only when the user has intentionally scrolled up from the bottom of the window. This is how many other LLM chat interfaces work. That way, if the current scroll position is at the bottom of the stream, then the auto-scroll will stay active because it's safe to assume the user wants to see the real-time stream. But if the user has moved the scroll position to a previous point in the stream, then the assumption should be that the user wants auto-scroll to be disabled.
JuliaBonita commentedon Feb 12, 2026
This issue does not seem to happen when using Kimi K2.5, but it happens constantly with Claude. So it appears this is caused by how OC is interpreting and rendering Claude's (and mabye other models) specific responses. I notice that Kimi K2.5 streams its responses a bit differently; so it should be easy to visually see the difference.
jcontini commentedon Feb 24, 2026
+1
Toggled scrollbars on, it still jumps down on tool completions. My eyes...they hurt
stee-re commentedon Mar 28, 2026
yes, I agree - this automatic scrolling down when the AI response changes makes this tool a little frustrating to use - I'm being forced to wait for some lengthy responses (so often a lot more than I wanted) until the Agent is completely finished. Then I can begin reading the response. It makes the process so much slower :-(
Otherwise - TOTALLY LOVING this tool - really really awesome work guys
John-Gee commentedon Apr 11, 2026
I'm having this issue with minimax, I'm not sure why the model should matter though here...
kreasteve commentedon Apr 18, 2026
Looked at the source with Claude — turns out you can just add
to
~/.config/opencode/tui.jsonso it looks like this (create the file if needed):{ "$schema": "https://opencode.ai/tui.json", "mouse": true }Restart opencode and it works.
Tested with MiniMax-M2.5 on macOS in iTerm2 — does exactly what's asked for in this issue: scroll up (wheel or PageUp) pauses sticky, scroll back to the bottom resumes it.
Edit: On closer testing I can't actually reproduce the bug without the file either —
mousedefaults totrueand the desired behavior already works. So this might not be the actual fix. Sorry for the noise.github-actions commentedon Jul 16, 2026
To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.
davispuh commentedon Aug 18, 2026
This has been annoying me so much. I just want to read and not having to keep scrolling back up all the time.
davispuh commentedon Aug 18, 2026
There's open duplicate #29094 and looks like will get solved with #40323
nullbio commentedon Aug 24, 2026
Insane that this is still an issue. Reminds me of Claude Code, they had this bug for, what, a whole year?
diegorodriguezv commentedon Oct 10, 2026
yes, please fix this