CanvasConvo: A Spatial Canvas for Branching LLM Conversations
Move over linear chats: non-linear AI interactions on a spatial canvas boost exploration.
Researchers from LMU Munich (Amin et al.) introduce CanvasConvo, a conversational interface that transforms linear LLM chat into a branching conversation tree embedded in a spatial canvas. Users can explore what-if scenarios by branching directly from any message, enabling parallel development of alternative directions. The interface integrates these branches with a familiar chat view, allowing seamless switching between linear and non-linear interaction. Features include timeline-based navigation to revisit past turns, automatic tagging and summarization of branches, and context-aware controls such as user-defined goals and reusable prompts to maintain continuity across sessions.
CanvasConvo was evaluated in a 5-7 day field study with 24 participants. Results showed that the non-linear structure supports exploratory workflows, helping users manage long-running conversations and compare alternatives. The work highlights how spatial, non-linear interfaces can improve ideation and analysis tasks with LLMs, addressing a key limitation of current chat-based systems. The paper (arXiv:2605.15848) is a step toward more flexible AI interaction design, potentially influencing future tools for researchers, writers, and analysts.
- CanvasConvo transforms linear chat into a spatial, branching conversation tree.
- Features include timeline navigation, auto-tagging, summarization, and context-aware goals/prompts.
- 24-participant field study (5-7 days) showed non-linear structures improve exploratory workflows.
Why It Matters
Moves LLM interfaces beyond linear chats, enabling richer exploration and comparison of ideas.