dsh-knowledge-base-local: A Local Knowledge Base DSH plugin for DeepSeek Harness

codfishyao/dsh-knowledge-base-local

Data & KnowledgeVerified
Listed on 2026-10-18
Page last updated 2026-10-26

A DSH plugin that gives DeepSeek Harness a local Markdown knowledge base with hybrid retrieval, PDF/Office import, an MCP server, and encrypted multi-user access.

dsh-knowledge-base-local is a DeepSeek Harness plugin that keeps your material as plain Markdown under any absolute path, readable by any model or editor. It targets scattered notes, weak retrieval and format lock-in, letting the agent search, read, import and tidy documents inside a session. It offers hybrid BM25 plus vector retrieval, three vector generation modes, PDF and Office text extraction, an inbox that ingests files automatically, an MCP server, per-document encryption and four access roles, with only Node built-ins at runtime.

How to Install

install
npmdsh plugin --profile web add dsh-knowledge-base-local
Category
Data & Knowledge
Platform
DSH Plugin
Author
codfishyao
Distribution
Plugin

dsh-knowledge-base-local Key Features

Markdown as model-readable dataHybrid BM25 + vector retrievalPDF/Office text extraction importMCP server for external clientsAES-256-GCM encryption, multi-user
Listed on dsh-plugin.org

dsh-knowledge-base-local Repo Summary

What Does It Do?

It is a DSH plugin for DeepSeek Harness that keeps your documents as plain Markdown in any absolute path you choose, so any model or editor can read them directly. It solves scattered notes, weak retrieval and format lock-in by letting the agent search, read, import and manage your material inside the harness. Core capabilities include hybrid retrieval, three vector modes, PDF and Office text extraction, an MCP server, at-rest encryption and multi-user roles, with zero runtime dependencies beyond Node built-ins.

Core Features

  • Hybrid retrieval: BM25 lexical matching combined with vector cosine similarity and title/substring bonuses, returning document id, line numbers and a snippet.
  • Six agent tools: search, read, list, add, import and manage, covering the full lifecycle from ingestion to cleanup.
  • Three vector modes: offline local TF-IDF feature hashing, model enrichment, and OpenAI-compatible embeddings that fall back to local vectors on failure.
  • Multi-format import: text, Markdown, code and web URLs, plus text extraction from PDF, DOCX, XLSX and PPTX, with an inbox folder that imports and updates files automatically.
  • Security and collaboration: per-document AES-256-GCM encryption, four roles from owner to reader with per-collection scoping, off by default for single users.
  • Versions and trash: deletion is soft and recoverable, and every edit keeps a version history (20 by default).

How to Use This Plugin?

After enabling it in DSH, open the knowledge base panel in settings and pick a folder, or simply tell the agent to set the knowledge base to a given path; the setting lives in the user data directory and follows you across projects. Then use the agent tools or the CLI to search, add, import and tidy content. Files dropped into the inbox folder are ingested automatically, and an MCP server can expose the library to external clients when needed.

How to Troubleshoot This Plugin?

When vector generation fails, API mode falls back to local vectors so retrieval keeps working; check the embedding endpoint configuration as described in the official docs. Encrypted documents cannot be read while locked, so unlock first, and if access is denied verify that your role and collection scope cover the target document.

This page is an independent rewrite of the plugin's official README — for authoritative documentation and the latest changes, refer to the source: codfishyao/dsh-knowledge-base-local. The plugin is third-party code that runs on your machine once installed; inclusion does not imply endorsement — please review the source before installing.

Install, update, and uninstall this plugin in the Plugin Market of DeepSeek Harness's DSH Plugin Hub

DSH Plugin FAQ