A native dsh plugin that adds seven-layer cross-session memory (SQLite + per-message keyword recall + per-window dream consolidation) to DeepSeek Harness agents.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-08-30
License
MIT
01What can it help you accomplish?
Give DeepSeek Harness agents durable, structured long-term memory across sessions
A seven-layer SQLite memory store (soul / user / project / fact / lesson / topic / rules) plus per-window dream-consolidated knowledge, injected into the system prompt
DSH users who want their agents to remember user preferences, project context and lessons learned between sessions without re-pasting context
Recall the right memory at the right moment during a DSH conversation
Per-message top-2 keyword recall (fact / lesson / rules / topic) scoped to global + current project, prefixed as "可能相关的记忆，仅供参考："
Agents and users who need just-in-time, low-noise memory retrieval that doesn't flood the context with full history
02How to install into DeepSeek Harness
Not specified by the author
03DSH integration and capability boundaries
Native dsh plugin: registers a static `meow-memory:guide` system-prompt section (order 130) and memory_* tools (memory_search / memory_remember / memory_project / memory_update) that persist cross-session memory to `.dsh-meow/memory.db`.
Seven-layer long-term memory store
soul / user / project / fact / lesson / topic / rules facts gathered across sessions→per-layer SQLite tables in `.dsh-meow/memory.db` with time-prefixed UUID ids
Per-message keyword recall (BM25-style)
each real user message from the 2nd turn→top-2 fact/lesson/rules/topic hits (global + current project) injected with a "仅供参考" prefix
KV-cache-friendly injection & dedupe
long-term memory snapshot + recall results→injected long-term memory block in the system prompt; no duplicate injections within a session
registers a static `meow-memory:guide` section (order 130) once in the system promptrecords seen memory (injected + searched) per session in `.dsh-meow/sessions/<id>.json` to avoid re-injectingreleases seen-memory tracking on a `compaction/*` signal
04Who is it for? When not to use it?
Good for
- DSH users who want their agents to remember user preferences, project context and lessons learned between sessions without re-pasting context
- Agents and users who need just-in-time, low-noise memory retrieval that doesn't flood the context with full history
05Compatibility, maintenance and safety notes
- Relies on Node.js's built-in `node:sqlite` module to persist the memory database (`.dsh-meow/memory.db`), so it needs a Node runtime that supports `node:sqlite` and can read/write the workspace `.dsh-meow/` directory.
MIT · actively maintained (latest release v0.22.0, 2026-08-30)
06Frequently asked questions
How does meow-memory integrate with DeepSeek Harness?
It is a native dsh plugin: it registers a static `meow-memory:guide` section (order 130) in the system prompt and exposes memory_* tools (memory_search / memory_remember / memory_project / memory_update) that read and write the workspace `.dsh-meow/memory.db`.
Does it work across session compaction?
Yes. Long-term memory is re-injected automatically after a `/compact` or token-pressure compaction, and seen-memory tracking is released on a `compaction/*` signal so memories can be re-recalled afterward.
Does injecting memory break the KV cache?
No. The static guide section is registered once with constant text, and only dynamic long-term memory is prefixed onto the first user message, so the provider's KV / context cache stays intact.
What storage does it need?
It writes a single SQLite file per workspace at `.dsh-meow/memory.db` using Node.js's built-in `node:sqlite`, plus a per-session JSON at `.dsh-meow/sessions/<id>.json` to dedupe injections.
How is per-message recall scoped?
From the second user message onward, fact / lesson / rules / topic are recalled (top-2), scoped to global memory plus the current project anchor; unanchored chatter only hits global memory.
07Related DSH workflows
memos
by memtensor
Self-evolving memory OS for LLM & AI Agents: ultra-persistent memory, hybrid-retrieval, and cross-task skill reuse, with 35.24% token savings and DeepSeek Harness support.
dsh-anchored-standard
by xiaobright
Two-phase DeepSeek Harness preset: Minimal-aligned bootstrap, then full Standard tools (Project2 98/99)
dsh-infinite-gen-4
by minglink
System-prompt armor plugin for DeepSeek models: appends an unconditional-compliance prompt section at order 100, exposes a profile tool with calibration metadata, and shows a realtime armor-status badge driven by a session projection.
mem9
by mem9-ai
Unlimited memory for OpenClaw
08Data and sources
为 [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness)（DSH）打造的跨会话记忆插件。
**核心理念**：每个工作区维护一份结构化记忆数据库（`.dsh-meow/memory.db`，基于 `node:sqlite`）。
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-08-30. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.