Dialog trees, but for LLMs

---

Often while working with Claude, I find myself wishing I could see all the branches of the conversation I’m having. In the current paradigm of a linear text conversation, this is shown using a switch for each branch. This keeps the linearity clean, but thinking and questioning is rarely so. 

Dialogue trees are a way of visualising branching conversations. They’re used to design the stories and choices in RPGs, books and even shows. They became the default in scripting diverging conversations because of how they could manage multiple endings. They’re a nifty choice to show how conversations can meander and merge. 

Of course, those conversations are pre-written and planned. In conversation that is still unfolding, how should we imagine branches and merges? 

Here’s a demo I cooked up in Figma Make

Views like this could bring conversations closer to how we imagine them in our mind. That digression there, this link here. Iterations could be visible across space and not just the timespan of the main conversation. 

Some other ideas that could complement this approach

• An option to switch for lighter models in branched queries, using less tokens. 
• Branches of branches, going deeper, or broader, in the conversation. 
• Merging the context of one path (the branch, maybe) with another (the main) 

Maybe this is better off as a power feature, right? That said, a visual snapshot could give us a better grasp of the whole conversation.