Hermes Pulpie cleans web pages locally for Hermes Agent
Hermes Pulpie is a plugin that changes how extracts useful text from web pages. It uses Pulpie, a 210 million parameter , to remove navigation menus, ads, sidebars, and footers from HTML and turn the remaining content into Markdown. It completes the cleanup in one model pass.
The stated quality is close to Dripper, with a ROUGE-5 F1 score of 0.862 versus Dripper’s 0.864, while using about one third of Dripper’s model size. It automatically uses a GPU through CUDA or MPS when available, and falls back to CPU when not. Typical pages are said to use about 80% fewer tokens after .
It , so it does not need an API key or a cloud service. Setup means installing the kachook/hermes-pulpie package with pip, then setting ’s web.extract_backend option to pulpie.
Key points
- can use Pulpie as its ion backend.
- It removes page clutter such as ads, menus, sidebars, and footers from HTML.
- It outputs cleaner Markdown for the agent to read.
- It claims about 80% token savings on typical pages.
- It without an API key or cloud .