That would be so nice. It would probs save people tokens. Have a lower model to do all the non reasoning task and use the big boys for complexity.
Indigo Sandwich
Yeah, that’s what I would use it for too. For me I could easily get the content of a pdf using gemini flash and then use a bigger model for the real prompt.
Salmon Jaguar
cool idea, but what if it worked retroactively like in chatgpt, you can edit previous messages and then choose different models while editing, then entering will make a new branch with current model (or the same model if you just want to reroll)
Indigo Sandwich
I think this should be done by just saving chats, if you go to edit one it appears as a branch, and the old ones dont dissapear. I like that one better as it has more control.
Purple Cupcake
I totally agree. I made this feature request which is kind of the same thing, but I didn’t feel like it was exactly a “duplicate”. Hopefully some elegant solution is put into place because the bad edit handling is one of the top things holding me back from going all-in on this platform.
I think this implementation needs a bit rework. This version can be useful for some usecases but; IMO branching, does not meant to be a new chat always. It would be good if we can see branches for our edits. When I edit something, I want to compare my previous for my questions. I like the chatgpt approach on this one.
Plus, there can be a tree view for better ux that differs t3chat from chatgpt. We can see chat branching on the side click of a button and select a chat to message to continue with etc. This feature may be an unnecessary feature conflicting with the main idea of t3chat but it would be useful IMHO.
Yellow Nucleon
This would be great, or to select some models and pick the response you’d like
Blue Raindrop
Really confused why this isnt a feature yet
Copper Canary
Probably too much but visual tool like obsidian graph would go hard with this.. :)
Log in to comment and vote
Comments22
Surprise! We just shipped branching
🫡
Turquoise Skybase
That would be so nice. It would probs save people tokens. Have a lower model to do all the non reasoning task and use the big boys for complexity.
Indigo Sandwich
Yeah, that’s what I would use it for too. For me I could easily get the content of a pdf using gemini flash and then use a bigger model for the real prompt.
Salmon Jaguar
cool idea, but what if it worked retroactively like in chatgpt, you can edit previous messages and then choose different models while editing, then entering will make a new branch with current model (or the same model if you just want to reroll)
Indigo Sandwich
I think this should be done by just saving chats, if you go to edit one it appears as a branch, and the old ones dont dissapear. I like that one better as it has more control.
Purple Cupcake
I totally agree. I made this feature request which is kind of the same thing, but I didn’t feel like it was exactly a “duplicate”. Hopefully some elegant solution is put into place because the bad edit handling is one of the top things holding me back from going all-in on this platform.
https://feedback.t3.chat/p/message-edit-history
Coffee Cactus
https://big-agi.com/blog/beam-multi-model-ai-reasoning
Here is an example of how this could look. Really awesome feature.
Ivory Control
I think this implementation needs a bit rework. This version can be useful for some usecases but; IMO branching, does not meant to be a new chat always. It would be good if we can see branches for our edits. When I edit something, I want to compare my previous for my questions. I like the chatgpt approach on this one.
Plus, there can be a tree view for better ux that differs t3chat from chatgpt. We can see chat branching on the side click of a button and select a chat to message to continue with etc. This feature may be an unnecessary feature conflicting with the main idea of t3chat but it would be useful IMHO.
Yellow Nucleon
This would be great, or to select some models and pick the response you’d like
Blue Raindrop
Really confused why this isnt a feature yet
Copper Canary
Probably too much but visual tool like obsidian graph would go hard with this.. :)