DeepSeek Harness (DSH) often faces context loss when handling multi-turn conversations. When a project involves multiple Agents or sessions, if no persistence is applied, previous conversation records and confirmed facts cannot be reused in the next session. The dsh-workspace-memory plugin solves this issue by introducing a workspace-isolated long-term memory mechanism.

Core Features

  • Workspace isolation: Memory for different projects does not interfere with each other, while global preferences can be shared across projects.
  • Multi-Agent support: Supports standard Agents, background Agents, and voice Agents to use the same project memory.
  • Memory management: Provides retrieval, explicit writing, deletion, conflict correction, and auditable historical versions.
  • Automation: No additional deployment of vector databases or embedding services is required. It automatically buffers, organizes, and retrieves relevant memories in subsequent tasks.
  • Command support: Supports commands such as memory_search, memory_remember, and memory_forget.

Installation and Activation

Before installing, ensure the environment meets the following requirements:

  1. DeepSeek Harness is installed.
  2. Use the DSH web profile.
  3. Node.js version >= 22.19.0.

Run the following command to install:

dsh plugin --profile web add --config.minimumReleaseAge=0 @flowingspring/dsh-workspace-memory@0.2.15

After installation, restart dsh web. After restarting, you can view and manage memories on the “Memory” page in the settings.

Usage

After successful installation, use the following command to verify the plugin version:

dsh plugin --profile web list

In conversations, Agents automatically retrieve relevant memories. You can also explicitly use commands:

  • memory_search: Perform a secondary retrieval.
  • memory_remember: Explicitly save a long-term fact.
  • memory_forget: Delete an entry by memory ID.

Use Cases and Notes

This plugin is suitable for the following scenarios:

  • You want multiple Sessions of the same project to share context.
  • Standard Agents, background Agents, and voice Agents need to use the same project memory.
  • Long-term memories need to be viewable, correctable, and deletable rather than a black box.
  • You do not want to deploy additional vector databases or embedding services.

Notes:

  • Uninstalling the plugin does not automatically delete saved memory data. You must manually delete the data directory.
  • Credential-type content is rejected from persistence and is re-masked before memory output.
  • After deleting a project, memories move to the “Recycle Bin” on the settings page and must be manually and permanently deleted.
  • The default data directory is $DSH_HOME/workspace-memory or ~/.dsh/workspace-memory.

Summary

This plugin reduces the context maintenance cost of Agent development through a structured long-term memory mechanism.

  • GitHub: https://github.com/lgquan/dsh-workspace-memory
  • Directory page: https://www.skillhub.cn/plugins/lgquan/dsh-workspace-memory