Introduction

In DeepSeek Harness (DSH) development practice, agents often rely on conversation history or static skill documents to obtain context. These approaches usually cannot durably store the “facts” or “notes” that agents want to remember, especially across sessions. As a community plugin, dsh-memory fills this gap by providing agents with persistent, editable storage capabilities.

Plugin Overview

dsh-memory is a community plugin that provides DeepSeek Harness with persistent, editable memory/notes storage. It allows agents to write, read, and search for facts as stable key-value pairs that remain available after a session ends. The plugin is maintained by Amengclass, uses the MIT License, requires no build step, and runs directly as pure ESM JavaScript.

Core Features

The plugin provides four core tools for managing an agent’s long-term memory:

  1. memory_set: Persistently inserts or updates facts/notes under a stable key (supports optional tags).
  2. memory_get: Reads a single entry by key; if the key is omitted, lists all entries.
  3. memory_delete: Deletes an entry by key (idempotent operation).
  4. memory_search: Performs case-insensitive substring search across keys, values, and tags, with support for filtering by required tags.

Installation and Enablement

Install the plugin into a Harness profile configuration. Since the plugin is distributed as pure ESM JavaScript, no build step is required:

dsh plugin --profile web add github:Amengclass/dsh-memory

The plugin only declares the storage domain; the storage backend is provided by the deployment environment (for example, the web profile defaults to @deepseek-ai/dsh-storage-json), so no backend configuration changes are needed.

Typical Usage

Agents can manage memory by calling the tools above. For example, set a convention in a project:

memory_set key="project/convention" value="Do not modify public/" tags=["project","rule"]

In a later session, the agent can retrieve the entry:

memory_get key="project/convention"

Use Cases and Notes

This plugin is suitable for scenarios where agents need to maintain specific state or a knowledge base across multiple sessions.

  • Dependencies: Runtime peer dependencies are required, usually provided by Harness’s node_modules: @deepseek-ai/dsh-storage-domain, @deepseek-ai/dsh-tools, @deepseek-ai/schemastery, and zod.
  • Storage Mechanism: Data is stored in the deployment configuration’s /sqlite or JSON backend, without involving manual file operations.
  • License: MIT License, permitting free use and modification.

Conclusion

dsh-memory provides DSH agents with a foundational persistent knowledge layer, enabling them to retain long-term memory much like humans. For more details, refer to the GitHub repository or the community directory.