dsh-memory-panel

Every time a conversation ends, an AI agent forgets everything—project conventions, user preferences, past decisions—back to zero. dsh-memory-panel equips every DeepSeek Harness agent session with durable, restart-proof memory, so context no longer lives and dies with the current session.

Core capabilities revolve around five model-facing tools: memory_save to create or update entries, memory_search for ranked keyword retrieval, memory_list for tag-filtered listing with stats, memory_get to read a single full entry, and memory_delete to remove one by id. On the system side, the memory:recall mechanism automatically injects the six highest-importance memories into every prompt assembly, letting agents carry context forward without repeated user reminders. Scoring weights titles, tags, and content at different multipliers, applies CJK bigram overlap for Chinese queries, and boosts high-importance entries even on no-hit searches.

It is built for day-to-day DSH developers who want their agents to genuinely remember: you simply talk naturally, and the agent saves key information at the right moments and recalls it in subsequent sessions. The Settings panel lets you browse, search, edit, or delete every memory entry at a glance.