omdsh-dev/dsh-mnemon
Cross-agent, local-first persistent memory plugin for DeepSeek Harness (DSH), powered by Mnemon. It shares long-term memory across Mnemon-enabled agents and adds runtime memory, searchable project documents, semantic recall, knowledge graph, and a Sidebar UI.
Install
via npm
Resolves a published tarball rather than cloning the repository, and installs without any extra setup. Swap `web` for your profile name if you run another one.
via GitHub · npm package
Installing from GitHub runs the project's build script, which pnpm blocks until you allowlist it — run the command once and pnpm prints the exact key to add under `allowBuilds` in ~/.dsh/profiles/web/pnpm-workspace.yaml.
What happened when we ran it
Installed cleanly when we ran it
Every command here is run in a throwaway container against a clean profile, and the result is whatever the harness recorded — not a guess from the source. Last run yesterday.
Due diligence
What it is — DeepSeek Harness plugin integrating Mnemon for persistent memory and cross-agent sharing.
Who it is for — When using DeepSeek Harness for multi-turn document-heavy tasks, people who need persistent context are suitable to install this plugin. When a task only relies on real-time tool results without needing historical memory, it does not need to install this plugin.
Watch out — Sandbox test passed in a fresh profile. Static check indicates execution of shell commands. No obvious pitfalls found.
The verdict — I would install it because the sandbox test passed providing persistent memory support, but the premise is first installing Mnemon, and only worth when cross-agent sharing is needed.
Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.
dsh-mnemon is a DeepSeek Harness plugin that adds local-first, persistent memory shared across Mnemon-enabled agents. It separates information into Runtime Memory, Project Documents, and Memory Spaces rather than copying everything into one store. Runtime preferences, collaboration rules, project conventions, and environment facts are projected as compact USER.md / MEMORY.md context on each turn. Documents remain complete Markdown references that can be searched before being loaded. Memory Spaces provide bounded recall for cross-session facts, decisions, entities, and relationships.
The deterministic dsh-mnemon Host handles storage, workspace and interaction rules. Search performs provider-native, read-only recall; Agent Query gives a clean task Agent bounded evidence; Remember and Save to memory ask a task Agent to qualify, route, deduplicate and distil accepted writes. The native Mnemon engine uses its local CLI and SQLite. Other supported providers include OpenViking, Honcho, Mem0, Hindsight, Holographic, RetainDB, ByteRover and Supermemory. Settings → Memory System → Background task Agent selects the provider and model for background tasks.
This suits agents that need durable, searchable context across sessions or shared long-term memory. It is a poor fit where transient conversation context is sufficient, or where adding a separate memory service is undesirable. Mnemon must be installed separately, and the README requires Node.js 20 or newer. The terminal surface is relevant because local providers can be reached through CLI programs, including Mnemon and ByteRover, with the agent operating under its available permissions.
The plugin exposes three non-interchangeable tiers:
| Tier | Content | Access |
|---|---|---|
Runtime |
Preferences, rules, conventions and environment facts | Compact USER.md / MEMORY.md projection on every turn |
Documents |
Designs, investigations, procedures, postmortems and handoffs | Search first, then full Markdown on demand |
Memory Spaces |
Cross-session facts, decisions, entities and relations | Bounded recall from active spaces |
Current instructions, repository files and live tool results take priority over historical memory. Background tasks use DSH’s new-session model route by default and do not reuse or consume the main conversation history.
The WebUI actions are deliberately separated by data effect:
Search performs concurrent, provider-native read-only recall.Agent query starts a clean top-level task Agent with bounded evidence and returns an answer without writing memory.Remember and Save to memory let a task Agent qualify, route, deduplicate and distil a candidate; the Host writes only accepted results.Smart selection applies hard provider rules and asks a task Agent to resolve genuine ambiguity, then saves a routing receipt.AI metadata creates local titles and descriptions asynchronously for selected Memory Spaces.Archive Document creates a searchable cold reference before the Host moves the original.Turn memory expands exact recall, write and Document-search activity, with navigation to each source.Mnemon is the official prioritised native engine, using a local CLI and SQLite. Optional providers are OpenViking, Honcho, Mem0, Hindsight, Holographic, RetainDB, ByteRover and Supermemory. External providers are off by default. Provider capability differences remain visible; the plugin does not invent graph edges, deletion semantics or enumerable content.
Reusable Provider services are managed in Settings; concrete Memory Space instances, activation, scope and metadata are managed in Memory Spaces. Settings → Memory System → Background task Agent selects a dedicated Provider and model for background tasks. The README requires Node.js 20 or newer and a separately installed Mnemon. macOS and Linux installation paths use Homebrew or Go; Windows uses the official v0.2.3-or-newer release ZIP.
Written from the project's own documentation and kept in sync with it. Where the two disagree, the source is authoritative — read the README on GitHub
Same category
volcengine
OpenViking memory and context bundle for DeepSeek Harness: pre-step auto-recall and profile injection, session capture, `viking://` URI guarding, and recall/write memory tools backed by an OpenViking server.
No one-line install — this plugin lives inside a larger repository and publishes no npm package.
vectorize-io
Hindsight, agent memory that learns: long-term project memory with auto recall and retain, knowledge pages, deep reflection, and per-repo memory banks.
No one-line install — this plugin lives inside a larger repository and publishes no npm package.
bowenliang123
Context insight panel: see what the model's context window is made of and how it evolves — composition vs. window size, per-request history, compression/injection events, and per-message token stats.
Installed cleanly when we ran it
csyangwen
为 DeepSeek Harness 带来「跨会话长期记忆 + 后台自我进化」能力的纯插件实现：五轨记忆 · git 分支感知 · 回合内自我审查 · 技能自我进化与技能管理器 · 四轨待办 · COI 调度 · 会话广播 · 会话搜索 · 提示词管理器 · 临时信息便签——零核心修改、零运行时依赖，随装随用、卸载即净。
Installed cleanly when we ran it
tinqiao-oss
The Engramory curated-memory discipline as an installable plugin ([npm: dsh-engramory](https://www.npmjs.com/package/dsh-engramory)): a deterministic 200-line / 25 KB cap on the `MEMORY.md` index via `ctx.tools.guard()` — growth denied, a shrinking rewrite always passes — plus the protocol registered as a runtime skill. The store is plain markdown, one file per fact, shared with Claude Code, Codex, Kiro, and OpenClaw.
Installed cleanly when we ran it
ZSeven-W
Noema long-term memory plugin for DSH: durable, inspectable agent memory with recall tools and a settings page.
Installed cleanly when we ran it