dsh-expmem

DSH ExpMem addresses the fundamental limitation where large language models lose personal experiences when conversations end or contexts are compacted. This plugin introduces a long-term memory architecture for DeepSeek Harness, separating ephemeral session history (the Recall layer) from distilled, persistent insights (the ExpMem archive) to ensure valuable knowledge is preserved and reusable across interactions.

The plugin's core capabilities revolve around intelligent memory management. It features tiered storage with pressure-aware promotion that prompts the Agent to preserve crucial information before context windows fill; personalized retrieval that ranks memories by relevance, recency, and assigned importance; auditable reflection processes where higher-level insights cite source memories to form reasoning trees; and a complete trust lifecycle (candidate, verified, disputed, superseded) that maintains provenance and reliability.

DSH ExpMem is ideal for power users who need to persist work patterns—developers executing recurring tasks, engineers accumulating debugging insights, or researchers studying long-term Agent behaviors. Operating as lightweight local files without requiring vector databases or background services, it enables any DeepSeek Harness user to build a personal, searchable, and verifiable experience library.