AI Agent Hub
Back to plugins
🧰

dsh-network

Web Tools Updated 2026.09.12

Run the following command in DeepSeek Harness:

dsh plugin install NaivG/dsh-network

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install NaivG/dsh-network in your terminal to install; the source is available at https://github.com/NaivG/dsh-network

About this plugin

LLMs need to search the web, fetch pages, and digest documents mid-conversation, yet most harnesses offer little built-in support. dsh-network plugs a complete web toolchain into DeepSeek Harness so the model can search, fetch, and process external content natively without an external proxy or scraper pipeline.

Core capabilities include web_search with cited sources and confidence markers, web_fetch that auto-detects PDF, Office, and EPUB formats and converts them to clean Markdown via officeparser, and server-side LRU caching with cacheId-based paging for responses over 20 KB. A low-level http_request tool offers full method, header, and body control, while web_config exposes live read and patch operations. All traffic runs through undici bound to 127.0.0.1 only, with SSRF validation, IP pinning, and private-range blocking enabled by default; true binary content is rejected at the protocol level.

Ideal for developers and writers who want their model to independently look up papers, scrape API docs, or read Office files during a conversation, and for teams that prioritise secrets isolation, loopback-only binding, and minimal network exposure.

Use Cases

  • Model searches the public web mid-conversation and returns cited, structured summaries with confidence markers
  • Fetches any HTTPS URL and auto-parses PDF, Office, and EPUB into clean Markdown for the model
  • Pages through large cached responses via cacheId and offset instead of re-downloading

Best For

  • Developers who want their model to independently look up papers and API docs
  • Technical writers who expect the model to digest Office and PDF files natively
  • Security-conscious teams that prioritise loopback-only binding and secrets isolation