Introduction

By default, DeepSeek Harness (DSH) conversation sessions are stateless. When a session ends or the application restarts, context is lost. This can be a barrier in agent development scenarios that require long-term memory, specific naming, or privacy isolation. The dsh-memory plugin solves these issues through persistent storage.

What Is This

dsh-memory is a static profile plugin maintained by developer hmlyx under the MIT license. It gives AI in DSH persistent memory, with support for memory tab management, automatic naming, privacy isolation across multiple AIs, and automatic archiving of experience.

Core Features

The plugin provides the following core capabilities:

  • Memory Tab: Adds a Memory tab in the conversation view, with support for a master switch, renaming, and file browsing.
  • Automatic Naming: Includes a built-in library of 50 unique AI names that can be assigned randomly.
  • Privacy Isolation: Records are divided into private (global, brief) and shared (shared, experience) modes to ensure data isolation between different AIs.
  • Sidebar AI Name: With one click, display each AI’s name in all session sidebars.
  • Experience Sharing: Share previously solved problems in read-only mode to enable reuse of experience.
  • Memory Archiving: Provides the memory_record tool, supporting archiving content as one of five types: global, shared, brief, event, or experience.

Installation and Activation

The plugin is installed as a static profile and does not require complex configuration steps.

  1. Place the repository
    Put the plugin repository into the DSH profiles directory:
    ~\.dsh\profiles\web\node_modules\dsh-memory\

  2. Update the dependency configuration
    Open ~\.dsh\profiles\web\package.json and add the following to the dependencies field:

    "dsh-memory": "3.0.0"
  1. Register the plugin
    Add the following to the dsh.profile.bundles field:
    "dsh-memory"
  1. Restart
    Restart DeepSeek Harness to apply the configuration.

Typical Usage

The plugin guides AI memory management by injecting tools and prompts.

1. Use Memory Tools
AI can invoke the memory_record tool to write memories. The kind parameter determines the type and visibility of the memory:
* global: A global private record.
* shared: A globally shared record (readable by others).
* brief: A brief private record.
* event: An event private record.
* experience: A shared experience record (a solved problem that can be reused).

2. Proactively Record Experience
* When solving a problem or learning a technique, proactively invoke the tool to record experience.
* Before completing a task or starting a new task, summarize an experience with a title.

3. SystemPrompt Injection
Inject the AI name and memory obligations into the systemPrompt without requiring user reminders.

4. Title Restoration
The plugin automatically manages sidebar names. When a session ends, the sidebar name is automatically restored to the original title.

Use Cases and Notes

  • Use Cases: Developers who need long-term conversation context, want AI to have self-naming capabilities, or need to isolate privacy between different presets.
  • Permission Notes: The plugin runs with the permissions of the current DSH process. Ensure the installation source is trustworthy.
  • Compatibility: Compatible with older dynamic plugins; data will not be migrated or lost. Memory files are stored by default in the conversation memory directory, and the state can be checked via .memory-state.json.

Conclusion

With dsh-memory, DSH users can build agents with long-term memory and privacy isolation capabilities. The plugin takes effect after installation and restart, and is suitable for use with any preset.

GitHub Repository
Skill Hub Directory Page