Research workbenches

dsh-library

Turn local Markdown and text files into a library the agent can cite: import, hybrid search that returns numbered sources, and a checker for the [n] markers in an answer. Markdown and text only; it does not read PDFs.

PerryLink

Install

  1. Click Plugins in the sidebar, then Add plugin at the top right
  2. Enter the package name dsh-library, press Install
  3. Restart when asked
dsh plugin --profile web add dsh-library
  1. Restart dsh web
  2. No dsh yet? npm i -g @deepseek-ai/dsh, needs Node 22.19+

What the code shows

Six tools. library_add chunks and indexes a file; library_remove deletes it and then probes the remaining index for residue; library_search ranks by keywords and vectors together, re-ranks for diversity and pins the strongest chunks to the head and tail of the result; library_cite_check checks each [n] marker in an answer against the result page; library_diagnose reports chunk sizes, near-duplicates and a self-retrieval probe; library_list shows what is indexed. The index lives in dsh's own storage. The default embedder is deterministic hashing, so nothing is downloaded, but a paraphrase that shares no words with the source will not be found and keywords do most of the work. Optionally it can use a local Ollama server (127.0.0.1:11434) or an embedder command you configure; it contacts nothing else. It imports Markdown and text only, so PDFs have to be converted first. Its only runtime dependency is zod. Apache-2.0. Maintained by PerryLink, not by us.

Listing is not endorsement. A plugin runs with your permissions: look at its source before installing.

Related