Cymose

Overview

Cymose breaks away from traditional, linear AI chat flows by introducing a tree-structured conversation environment. Instead of stuffing a single linear chat history with unrelated tangents or forcing you to jump between completely fragmented chats, Cymose organizes your interaction into hierarchical workspaces.

It uses a hybrid model combining a highly optimized native engine and a cross-platform presentation layer to maximize privacy, speed, and token efficiency.

Core Concepts

• Zero-Token Isolation: When you branch out and create a child workspace, it starts with an entirely blank message log. The AI is not bogged down by the raw transcript of your sibling tangents, preventing context pollution and saving token costs.

• Hierarchical Memory Cascade (ancestor_trail): A child workspace automatically inherits the condensed context of all its parent workspaces. This ensures the model retains the high-level project vision or core parameters without needing to reread the whole linear log.

• Eager Upward Propagation (descendant_digest): When you commit a summary or key breakthrough deep down in a sub-workspace, that information propagates back up to the root instantly. The effective summary cache is updated in O(depth) time upon writing, making cross-branch discoveries immediately available from the root node.