YYTbit/dsh-plugin-meta-memory

Structured long-term memory system for DeepSeek Harness

Structured long-term memory system: teaches the agent to organize memories into Units (work categories named verb-modifier-noun.unit), Records (each memory has a brief for quick scan and a full version for detail), a self-maintained global index, and auto-injection of relevant briefs into context (enableInjection). Includes an mm CLI (init/list/read/search/index). Briefs are capped by maxBriefBytes so context stays lean; full records load on demand.

Agent Capabilities ★ 3 updated 2026-08-14 — untested
View on GitHub ↗

Install

dsh plugin --profile your-profile add dsh-plugin-meta-memory

npm package dsh-plugin-meta-memory 0.1.1 (registry-verified 2026-08-26). dsh plugin --profile your-profile add dsh-plugin-meta-memory. Configuration via the plugin row (memoryPath default ~/.dsh/meta-memory, enableInjection, enableSkill, maxBriefBytes). MIT.

Compatibility

DeepSeek Harness (any profile). Structured memory layout under ~/.dsh/meta-memory with units/records/index.

Details

Recent updates

v0.1.1: units/records/index structure; auto-injection; mm CLI; maxBriefBytes cap.

FAQ

How is memory organized?
Units (work categories), records with brief + full versions, and a self-maintained global index under ~/.dsh/meta-memory.
What gets injected into context?
Relevant briefs only (bounded by maxBriefBytes); full records load on demand via mm read.
How do I manage it?
The mm CLI provides init, list, read, search, and index commands.

Alternatives

csyangwen/dsh-memory-evolve · KLRSL/dsh-biomemory · Aik358/dsh-auto-memory

More plugins in Agent Capabilities

Browse more in Agent Capabilities

Guides for Agent Capabilities plugins