AI Agent Hub
Back to plugins
dsh-memory preview

dsh-memory

Memory Updated 2026.09.05

Run the following command in DeepSeek Harness:

dsh plugin install chenhw7/dsh-memory

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install chenhw7/dsh-memory in DeepSeek Harness to install; see the source at https://github.com/chenhw7/dsh-memory

About this plugin

Every time you close a session, the DeepSeek Harness agent forgets the facts, preferences, and hard-won lessons it gathered along the way, forcing you to re-explain context from scratch. Cairn (@chenhw7/dsh-memory) persists what matters into a local KV store and re-injects it in subsequent sessions, giving the agent genuine long-term memory that survives across sessions and restarts.\n\nUnder the hood it offers CJK-aware BM25 relevance search, three-layer scoping (global, project, user), nine model-facing memory tools, an automatic learning pipeline that captures remember-intents, corrections, and verified failure-recovery patterns from live conversation, write- and load-time security scanning that blocks API keys and prompt-injection attempts, a pin / soft-decay lifecycle, and an optional human-review queue before anything is written. All settings are exposed through four live-apply cards in the Settings UI.\n\nIt is built for developers and teams who collaborate across projects over time and do not want to re-explain coding preferences and pitfalls every session, as well as for anyone who wants a structured memory layer in their agent workflow while keeping all data fully local and under their control.

Screenshots

Use Cases

  • Retain coding preferences, project conventions, and pitfall logs across sessions without re-explaining
  • Automatically learn fix patterns and failure-recovery lessons from conversation into searchable entries
  • Share global preferences across projects while keeping per-repo context isolated

Best For

  • Individual developers using DSH for day-to-day coding
  • Engineering teams collaborating across repos with shared coding conventions
  • Users who want a structured agent memory layer with fully local data control