In the development workflow of DeepSeek Harness (DSH), debugging the state of the memory system (such as the distribution across Layer 0, 1, 2, and 3) is usually implicit. It is difficult for developers to intuitively confirm whether memories are captured or retrieved correctly. DSH uses a plugin-based architecture. The following section introduces a local memory browser embedded in the conversation view.

Plugin Positioning

dsh-memory-dashboard is a plugin maintained by perinchiang. It provides read-only browsing of the local four-layer memory directly in the Harness conversation interface. It does not replace the memory backend, nor does it write data to the Tencent Cloud gateway; it exists only as a data viewing tool.

Core Capabilities

  1. Embedded View
    The plugin is injected into Harness’s conversation.view and does not provide a standalone Dashboard page.

  2. Overview Statistics
    The top of the interface displays the counts for L0/L1/L2/L3, the current layer proportion, and session statistics.

  3. Interaction and Filtering
    The Layer Selector is linked with the record list and supports local search and loading more items.

  4. Detail Viewing
    Memory details use a Bottom Sheet on mobile, and technical metadata is collapsed by default. When a locatable Session exists, you can jump back from the details to the original Harness session.

  5. Interface Adaptation
    Desktop uses a centered Content Shell, while Mobile uses the same information architecture and avoids horizontal overflow.

  6. Technical Implementation
    The host opens SQLite in read-only mode; the browser accesses the plugin API through the current DSH Web port.

Installation and Configuration

Install the plugin using the official command:

dsh plugin --profile web add github:perinchiang/dsh-memory-dashboard

After installation, you need to restart dsh web. In production, it is recommended to pin the GitHub installation to a reviewed commit to ensure stability:

dsh plugin --profile web add github:perinchiang/dsh-memory-dashboard#<commit-sha>

By default, the plugin reads the SQLite database at the following path:

~/.memory-tencentdb/memory-tdai/vectors.db

If the data directory is not in the default location, you can specify it with an environment variable:

TDAI_DATA_DIR=C:\path\to\memory-tdai

Usage Steps

  1. Install and restart dsh web.
  2. Go to the conversation page.
  3. Open the “Memory” entry at the top of the page.

Use Cases and Considerations

This plugin depends on dsh-tdai-memory being installed and running locally, and the vectors.db file must exist locally.

The plugin runs on the host with read-only permissions and has the following characteristics:
* It does not call Tencent Cloud APIs and does not contain API keys.
* It does not modify database content.
* It runs only as a browser plugin and requires no TypeScript build step.

The design philosophy of DSH is “everything is a plugin”. This plugin is a project in the community directory and has no official affiliation with DeepSeek or Huanfang Research Institute.

Conclusion

dsh-memory-dashboard provides DSH developers with a visual window for inspecting the state of the memory system, making it a powerful auxiliary tool for debugging agent memory workflows.