Introduction

In the DeepSeek Harness plugin ecosystem, building long-term interactive scenarios often faces two problems: models tend to turn guesses into facts, or users need to maintain an additional memory management system. Mind Garden aims to address these issues. As an installable long-term companion for the DeepSeek Harness Web profile, it provides a private reflection space. Its core positioning is “a long-term companion that accurately understands you and allows you to correct it.”

What It Is

Mind Garden is a long-term companion and private reflection space maintained by hyyhf. The plugin reuses Harness’s providers, credentials, storage, attachments, remote features, and plugin lifecycle without forking or copying the DeepSeek Harness application. It provides a continuous conversational experience through encrypted personal records, user-controlled memory, nine interconnected reflection spaces, and cinematic 3D scenes.

Core Features

The plugin primarily provides the following capabilities:
* User-controlled memory: Users have control over memory, and the model does not own the user’s story.
* Star Observer: Provides the ability to draw and correct conversations.
* Visual feedback: Includes cinematic photo particles and living 3D constellations.
* Photo observation: Uses the official DeepSeek routing, supports the deepseek-v4-flash-vision-exp model, and requires a Harness-managed DEEPSEEK_API_KEY.
* Reflection spaces: Includes nine interconnected reflection spaces.

Installation and Activation

Before installing, ensure that DeepSeek Harness 0.1.1-rc.1 or a newer version is installed and that pnpm is available in the system path.

dsh plugin --profile web add git+https://github.com/hyyhf/mindgarden.git

Typical Usage

After installation, you can enter the garden by following these steps:

  1. Verify the installation:
    dsh plugin --profile web why @deepseek-ai/dsh-mind-garden
  1. Start the Web interface:
    dsh web
  1. Configure interaction: In the dialogue posture panel, select two dialogue postures and five supported intents.

Use Cases and Notes

  • Use cases: Users who need long-term companionship, value private records, and want clear control over AI memory and dialogue direction.
  • Environment requirements: The Photo observation feature requires a Harness-managed DEEPSEEK_API_KEY.
  • Output control: Mind Garden does not add an application-owned maxTokens by default; output length is controlled by the selected provider and model.
  • Source compliance: Check the source code and license before installing.

Conclusion

By reusing DSH’s underlying capabilities, Mind Garden provides a controllable, long-term, and visually expressive conversational environment. It is suitable for users who need deep reflection or long-term record-keeping in the Web profile.