dsh-goodmemory：为 DeepSeek Harness 增加自动跨会话记忆
dsh-goodmemory 是一个为 DeepSeek Harness (DSH) 提供跨会话长期记忆的插件，由 hjqcan 维护，采用 GoodMemory 作为后端。该插件旨在解决 DSH 会话间缺乏稳定记忆的问题，通过在模型步骤前自动召回相关记忆（注入为可识别的 user message），并在回合成功完成后写回用户和助手的有效对话文本，同时排除工具调用、推理过程及注入内容。 插件支持两种运行模式：默认的 managed 模式会自动启动 goodmemory-http-bridge sidecar，使用 SQLite 存储数据；external 模式则连接已有的 GoodMemory
Read Moredsh-layered-memory：DSH 跨会话长期记忆插件
dsh-layered-memory 是 DeepSeek Harness (DSH) 的跨会话长期记忆插件，采用 MIT 授权，旨在解决会话结束后上下文丢失问题。该插件将验证过的事实、任务经验和环境信息持久化存储于磁盘，支持跨会话检索。 核心功能包括 L1-L3 分层记忆、命名空间隔离、BM25 全文检索、内容去重及跨命名空间经验提升。插件提供 14 个记忆工具，如 `memory_write`（强制证据验证）、`memory_search`、`memory_index` 等，支持渐进式工具暴露。一致性维护依赖查重、手动更新及合并候选，系统不自动检测矛盾，且严禁物理丢弃已验证事实，仅允许压缩
Read MoreDSH 跨会话记忆插件 dsh-memory：YAML 存储、语义检索与两级注入
`dsh-memory` 是一款专为 DeepSeek Harness（dsh）开发的 MIT 许可跨会话记忆插件，旨在解决新会话中上下文丢失的问题，如项目偏好和历史教训需反复说明的情况。该插件通过 YAML 文件（默认 `memory.yaml`）存储记忆，提供 `remember`、`recall`、`view`、`list`、`forget`、`pin` 六个核心工具，实现记忆的保存、混合检索（关键词+本地 Ollama embedding）、查看及生命周期管理。 其主要特性包括：采用“标题优先”注入策略以减少 Token 占用；支持 `session`（固定注入）和 `auto`（按需
Read Moredsh-memory-vault: A Cross-Session Memory Vault for DeepSeek Harness
dsh-memory-vault is a cross-session memory plugin developed for DeepSeek Harness (DSH), designed to solve the problem that information such as project preferences, deployment constraints, and test commands is confined to a single conversation context and cannot be persistently reused. The plugin provides three core model tools — `memory_remember`, `memory_recall`, and `memory_forget` — and persists memory data in the DSH storage domain. Its main features include: automatically injecting recent memory entries at each system prompt assembly to maintain contextual continuity; scoring and ranking recall results based on tag and content matching; and providing a browser-based management entry through the Settings page, allowing users to intuitively browse, add, and delete memories. The plugin is simple to install, requiring only
Read Moredsh-biomemory: Cross-Session Memory Plug-in for DeepSeek Harness
dsh-biomemory is a cross-session bio-inspired memory plugin for DeepSeek Harness (DSH), designed to address the issue of context loss in multi-turn tasks. It expands memory from individual logs into a structured management system through hierarchical storage, approval gating, semantic retrieval, memory metabolism, and audit snapshots. Built on SQLite and a local embedding model (bge-small-zh-v1.5), the plugin supports exact keyword, semantic vector, and hybrid retrieval, and automatically degrades to keyword retrieval when the model is unavailable. Core capabilities include frozen snapshot injection to reduce preference-guessing costs, manual approval and demotion strategies for important memories, and half-life decay, reference consolidation, and conflict surfacing via the "Dream" mechanism. Additionally, the plugin provides complete audit logs, backup recovery, and single-entry rollback functionality, with support for importing from Markdown.
Read MoreEquip DeepSeek Harness with Readable and Editable Cross-Session Memory Using dsh-mneme
dsh-mneme is a DeepSeek Harness memory plugin maintained by modusensus, licensed under MIT. It uses native SQLite with editable Markdown for dual-writing cross-session memory, and offers autoDream background consolidation and optional offline semantic retrieval. This article verifies the installation commands, seven memory_* tools, configuration options, and notes based on the community directory page and GitHub README: the plugin runs with current dsh privileges, requires Node 24+, and has entity extraction and Sleep Mode disabled by default.
Read More