dsh-cue-bank

dsh-cue-bank aims to bridge the gap between Agent memory and human memory. While humans rely on multi-dimensional touchpoints to switch between tasks, many Agent memory systems are either raw logs or require explicit references, leading to context clutter. This plugin simulates that mechanism, allowing Agents to automatically build a cross-session "cue" memory bank that can recall historical details without user prompting during topic switches.

Its core capability lies in "automatic memory and active awakening." By detecting topic switches via keyword overlap or semantic vector matching, the plugin dynamically retrieves relevant events from global storage and injects them into the current context without polluting the persistent conversation history. Supporting both local keyword matching and cloud vector retrieval modes, it balances low cost with high semantic understanding, making it ideal for Agents requiring long-term memory, role consistency, or multi-turn complex dialogues.

For developers, this is a powerful tool to enhance the intelligence of Agents. It works out of the box with simple threshold settings to control memory sensitivity. Whether building assistants that need to remember user preferences or project backgrounds, or managing context links in multi-Agent collaboration, dsh-cue-bank enables Agents to naturally remember the past, focusing on the current interaction just like humans do.