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
feat(settings): ability to change default thinking level #384
when setting a thinking level on a thread and creating a new one it goes back to default
might be worth adding a dropdown menu on the settings to change the default thinking level.
@maria-rcks I was planning to implement this. Which one would you prefer, Thinking Level in the settings page, or simply persist the last selected Thinking Level in the composer?
I don't think there is a guarantee that each model will use the same "reasoning" enumerations, especially across providers. My recommendation would be to not add a new settings item but make sure t3code picks the same default that the harness uses and then persist that as the last selected default.
i would also like to set the default model, reasoning and fast mode.
My recommendation would be to not add a new settings item but make sure t3code picks the same default that the harness uses and then persist that as the last selected default.
i like the "same default as harness" approach. codex has config settings for model, model_reasoning_effort and service_tier. you could read and expose these options in the t3code settings under a "Codex settings" section.
Activity
Dastari commentedon Mar 8, 2026
This would be good. New chats constantly default to "High" where I normally want to run "Medium"
MatejSkarka commentedon Mar 9, 2026
+1 to this, I also want to set default "medium" and it is defaulting to "High" and also it is defaulting to "Fast mode" off
maskdotdev commentedon Mar 13, 2026
@maria-rcks I was planning to implement this. Which one would you prefer, Thinking Level in the settings page, or simply persist the last selected Thinking Level in the composer?
See Below:
t3code-persist-thinking.mp4
Dastari commentedon Mar 13, 2026
I don't think there is a guarantee that each model will use the same "reasoning" enumerations, especially across providers. My recommendation would be to not add a new settings item but make sure t3code picks the same default that the harness uses and then persist that as the last selected default.
maskdotdev commentedon Mar 13, 2026
My suggested approach would only apply to codex, I'm assuming a similar approach can be applied to other providers when they become available.
thameraltoimi commentedon Mar 15, 2026
i would also like to set the default model, reasoning and fast mode.
i like the "same default as harness" approach. codex has config settings for
model,model_reasoning_effortandservice_tier. you could read and expose these options in the t3code settings under a "Codex settings" section.maria-rcks commentedon Mar 15, 2026
adding a whole section is not something i would encourge for a feat as simple as this
could you test my pr? and see if it does what you want