dsh-memory-vault

Every DeepSeek Harness conversation starts from a blank slate. Project context explained last week, coding preferences confirmed over many rounds, technical decisions hammered out in long threads all vanish the moment you open a new window. dsh-memory-vault plugs that gap by giving DSH a persistent, tiered memory layer: the plain-text exchange of each conversation is automatically archived, then progressively compressed across short-term, mid-term, long-term, and permanent tiers, so critical information survives session boundaries.

Under the hood the plugin injects a full suite of memory tools into DSH: add, list, search, show, update, archive, restore, and delete memories; manage categories and tags; retrieve conversation history; promote permanent entries; and maintain a global preset slot. The AI calls these tools transparently during dialogue, so you simply say remember this or what did we discuss last time in plain language. Permanent entries are pinned to the current session and will not be silently evicted; per-tier retention counts and compression batch sizes are all tunable via config, and the local store lives at a fixed path independent of your working directory.

It is built for DSH users who push a project forward across many sessions and no longer want to paste context into every prompt, who would rather structure scattered decisions and preferences into a searchable tiered archive, and who accumulate team conventions or personal invariants that must ride along into every new conversation automatically via the global preset.