Branching a DeepSeek Harness conversation without losing the original thread

When an AI coding assistant suggests two approaches, I want to explore both without mixing them into one long chat. I maintain ThoughtDAG, and I’ve added its canvas as a plugin inside DeepSeek Harness. You can switch from chat to a graph, select a phrase in an answer, and explore it as a separate branch. The connections determine which upstream nodes go into the next canvas request. A discarded approach can stay visible without being included in that request. Editing the graph does not rewrite the original Harness session log. The plugin uses your configured Harness providers. It’s MIT-licensed and independently maintained, not an official DeepSeek plugin. Source and installation: github.com/chenxiachan/thoughtdag/tree/main/dsh For people using DSH: would branching in place help you compare approaches, or would you rather keep separate chat sessions?