Repolode
omdsh-dev

omdsh-dev/dsh-mnemon

Memory and context layer for the DeepSeek Harness agent, with layered runtime memory, searchable project documents, pluggable memory providers and self-evolving skills.

Stars
919
Forks
63
Open issues
26
Last commit
8 hours ago

What is it?

dsh-mnemon is a plugin for DeepSeek Harness (DSH) that manages what the agent knows across sessions. Runtime memory such as preferences and working facts is injected into every turn as compact USER.md and MEMORY.md files, while project documents and long-term memory spaces are searched only when a question needs them.

The Plugins page lets you pick a main strategy, such as the default Layered strategy or the General strategy, and switch optional enhancements without moving any data. Storage can be local, per workspace or in a chosen directory, with ZIP backup. Memory Spaces can be backed by the local Mnemon Native provider or one of several third-party providers such as Mem0, Honcho and OpenViking.

The Memory System page shows stored memories, project documents and a graph of entities, and lets you edit them. A Skills component, marked experimental, manages DSH SKILL.md files, reviews finished conversations for improvements, backtests drafts and installs skills from skills.sh and GitHub.

Key features

  • 01Runtime memory injected each turn as USER.md and MEMORY.md.
  • 02Project Documents searched and read on demand.
  • 03Memory Spaces with local Mnemon Native or eight third-party providers.
  • 04Strategy and enhancement switching on the Plugins page without moving data.
  • 05Skills management with versions, backtests and installs from skills.sh and GitHub.
  • 06Per-turn view of which documents and memories were read and written.

Use cases

  • Giving a DSH agent persistent preferences and project knowledge across sessions.
  • Searching a project's design docs and handoff notes during a conversation.
  • Keeping memory local while optionally connecting a managed memory service.
  • Developing custom memory sources or strategies as DSH plugins.

Quick start

  1. 1

    Install pnpm globally

    npm install --global pnpm
  2. 2

    Start the DSH web interface

    npx @deepseek-ai/dsh web
  3. 3

    In the page, go to Plugins → Add plugin, type dsh-mnemon, click Install, then Enable now

Things to know

  • Requires Node.js 22.19 or later.
  • The Skills component is marked experimental, and self-evolution is off until you enable it.
  • Memory Spaces need the Mnemon CLI installed or another Provider enabled before they work.

In collections

Topics

FAQ

What is dsh-mnemon?
Memory and context layer for the DeepSeek Harness agent, with layered runtime memory, searchable project documents, pluggable memory providers and self-evolving skills. dsh-mnemon is a plugin for DeepSeek Harness (DSH) that manages what the agent knows across sessions. Runtime memory such as preferences and working facts is injected into every turn as compact USER.md and MEMORY.md files, while project documents and long-term memory spaces are searched only when a question needs them.
Is dsh-mnemon free to use? What license does it have?
dsh-mnemon is open source under the MIT license. Check the license terms before using it in commercial products.
Is dsh-mnemon still maintained?
Actively maintained. The last commit to dsh-mnemon was on Nov 7, 2026.
What can I use dsh-mnemon for?
Giving a DSH agent persistent preferences and project knowledge across sessions. Searching a project's design docs and handoff notes during a conversation. Keeping memory local while optionally connecting a managed memory service. Developing custom memory sources or strategies as DSH plugins.
How do I install dsh-mnemon?
Install pnpm globally: npm install --global pnpm Start the DSH web interface: npx @deepseek-ai/dsh web In the page, go to Plugins → Add plugin, type dsh-mnemon, click Install, then Enable now
Where can I get dsh-mnemon?
The source code and installation instructions are on GitHub: https://github.com/omdsh-dev/dsh-mnemon

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects