Introduction

The plugin architecture of DeepSeek Harness (DSH) allows developers to extend the capabilities of agents. In terms of localized personal knowledge management and long-term memory, MemoKnow provides a lightweight solution. By using structured records and document snapshots, it helps agents maintain contextual memory and retrieve knowledge from imported documents.

What Is MemoKnow

MemoKnow is a local personal memory and knowledge management plugin for DeepSeek Harness, maintained by twoyoung91. It addresses how agents can extract persistent facts from chat history and how to manage a personal knowledge base. The plugin stores records in SQLite, snapshots original documents using SHA-256, combines FTS5 full-text search with optional sqlite-vec semantic retrieval, and provides a management panel inside the DSH interface.

Installation and Enabling

  1. Download the corresponding version’s .tgz file from GitHub Releases (for example, dsh-external-dsh-memoknow-0.2.0.tgz).
  2. Run the installation command in the command line to add it to the web profile:
dsh plugin --profile web add C:\path\to\dsh-external-dsh-memoknow-0.2.0.tgz
  1. Restart DSH (if it is running), and then open Settings → MemoKnow.

Core Features

  • Memory Review and Import: Provides an interface in DSH settings for reviewing suggestions, allowing you to inspect evidence, edit suggested memories, or approve/reject them.
  • Knowledge Base Management: Supports importing plain text, Markdown, Word, PDF, CSV, and Excel documents. Imported content is stored as immutable snapshots and supports searching and previewing extracted text.
  • Control and Configuration: Supports pausing automatic learning, setting a token budget, and checking pending sessions and daily usage.
  • Retrieval Capabilities: Uses Local FTS by default (no model required). Optionally enable Local CPU embeddings (which will download a model) or an OpenAI-compatible embedding service.
  • Memory Maintenance: Supports editing or permanently forgetting individual memory records.

How to Use

  1. Open DSH Settings → MemoKnow to access the knowledge base management page.
  2. Review memories: View the list of suggested memories, inspect their source sessions and evidence, and decide whether to approve them.
  3. Search and preview: Search imported documents in the knowledge base and preview the extracted text content.
  4. Control learning: Adjust learning policies on the settings page, such as pausing automatic memory, limiting token usage, or viewing pending sessions.

Notes

  • Unofficial Component: This is a standalone community plugin and is not an official DeepSeek Harness component.
  • Single-user and Local: It is designed for one person’s local DSH profile and does not support multi-user setups or cloud synchronization.
  • Secure Storage: It stores only the environment variable name for the API key and never saves the actual key value.
  • Data Structure: SQLite is the authoritative source of metadata and lifecycle state, while original document snapshots are immutable.
  • Compatibility Limitations: Embedded images are ignored; legacy Excel (.xls) is not supported; direct installation via github: is not supported; the xls format is not supported.
  • Managed Mode Limitation: In managed embedding mode, Settings API updates cannot override policies.

Data Storage

The default data root directory is $DSH_HOME/memoknow; if DSH_HOME is not set, it is ~/.dsh/memoknow.

memoknow.sqlite3
objects/sha256/aa/bb/<full-sha256>[.txt]
models/                         # Created only after enabling Local CPU embeddings