htcqp802/dsh-knowledge-base
General-purpose knowledge base for DSH: import md/txt/json/yml/docx/pdf, folder management, FTS5 full-text search (BM25), and a Web management UI.
dsh-knowledge-base turns documents into a searchable knowledge base for DeepSeek Harness agents: drop in md/txt/json/yml/docx/pdf files (no size limit), and they are parsed, chunked by headings/paragraphs (adjacent small chunks merged), and instantly searchable with FTS5 trigram + BM25 ranking (LIKE fallback for short queries). Agent-native tools kb_query / kb_import / kb_list / kb_update / kb_delete work directly in conversation, while a Knowledge Base tab in the Web UI adds drag-and-drop import, folder management like an OS file manager, category administration and search. Re-importing the same file upserts instead of duplicating.
Install
dsh plugin --profile web add dsh-knowledge-basenpm dsh-knowledge-base 0.1.5 verified 2026-09-03 (repository field → github.com/htcqp802/dsh-knowledge-base; README EN primary with 简体中文 / Español / 日本語 editions). Install: dsh plugin --profile web add dsh-knowledge-base.
Compatibility
DSH Web profile; Node >= 22.5; SQLite FTS5 (trigram tokenizer); PDF parsed in-process with pdfjs-dist (zero system deps).
Details
- Repo: htcqp802/dsh-knowledge-base
- Category: Tools & Capabilities
- Stars: 0
- Version: npm dsh-knowledge-base 0.1.5
- Last push: 2026-08-16
- First seen: 2026-08-16
Recent updates
0.1.5 current on npm (verified 2026-09-03).
FAQ
- Which file formats are supported?
- Markdown, txt, json, yml, docx and pdf — PDFs are parsed in-process with pdfjs-dist, so no system-level PDF tools are needed.
- How does the agent search it?
- Through kb_query and the other kb_* tools, backed by SQLite FTS5 (trigram tokenizer, so no Chinese segmentation is needed) with BM25 ranking.
- Does re-importing a file duplicate it?
- No — re-importing the same file overwrites it (upsert), so duplicates do not accumulate.
Alternatives
ICCuse/dsh-knowledge · Luke-Yong/dsh-plugin-knowledge-graph