The core philosophy of DSH is “everything is a plugin,” which means its extensibility is highly flexible. However, in real-world use, relying only on the current context window quickly hits limits or causes the AI to forget earlier parts of the conversation. dsh-memory-vault is a plugin designed for this purpose. It uses an external database to provide DSH with a function similar to human long-term memory.

Core Features

This plugin mainly solves memory gaps during long agent conversations. Its specific capabilities include:

  • Memory Management: Supports adding, viewing, searching, modifying, and deleting memories, as well as archiving and restoring them.
  • Categories and Statistics: Provides category management, tag management, and statistics.
  • Conversation Records: Automatically saves each complete conversation as a record, storing only user and assistant text and excluding tool-call details.
  • History Search: Searches historical conversation content specifically.
  • Hierarchical Memory: Manages memories using a layered structure:
    • Short-term: recent conversations.
    • Medium-term: compressed summaries.
    • Long-term: further compressed summaries.
    • Permanent: important content in the current session.
  • Global Presets: Global “preset/user memory” is stored separately and automatically included at the start of each conversation.
  • Automatic Management: Permanent memory bound to the current session is not discarded; when long-term memory is full, it automatically absorbs important content and discards the oldest entry.

Installation and Enablement

The most straightforward way to install the plugin is via the command line. Run the following command in your terminal:

dsh plugin --profile web-desktop add github:hjj588/dsh-memory-vault

After installation, restart the DSH desktop client for the change to take effect. After restarting, find the plugin under “Settings → Plugins → Manage”.

Usage

Once the plugin is running, it injects a set of tools into DSH, which the AI can call automatically during conversations. You do not need to manually write tool-call instructions; simply express your needs naturally in the conversation. Common use cases include:

  • “Remember…”: Adds information to the long-term memory store.
  • “Search memory…”: Retrieves stored memory content.
  • “What was said earlier…”: Looks up historical conversation records.

Configuration and Tuning

By default, short-term memory retains 20 raw messages, compresses 10 at a time, and medium-term and long-term memory each retain 20 summaries. If you need to adjust these parameters, edit the configuration file ~/.dsh/profiles/web-desktop/cordis.patch.yml.

Add or modify the config field under the memory-vault configuration entry:

- insert:
    - id: memory-vault
      name: 'dsh-memory-vault'
      config:
        shortTermLimit: 20
        shortTermCompressCount: 10
        mediumTermLimit: 20
        mediumTermCompressCount: 10
        longTermLimit: 20

Notes

  • Data Storage: All memory data is stored in ~/.dsh/memory/memory.db, independent of the current workspace path.
  • Data Content: The plugin stores only user and assistant text, not tool-call details.
  • Runtime Permissions: The plugin runs with the permissions of the DSH process. It is recommended to review the source code and license before installation.

Summary

For developers who need to maintain context over long periods, conduct multi-round conversations, or build long-term-memory agents, dsh-memory-vault is a practical auxiliary tool. For more details, see the GitHub repository or the DSH plugin directory.