pi-fetch

Efficient web fetch tool for Pi

Package details

extension

Install pi-fetch from npm and Pi will load the resources declared by the package manifest.

$ pi install npm:pi-fetch

Pi manifest JSON

{
  "extensions": [
    "./extensions/fetch.ts"
  ]
}

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

Efficient web fetch tool for Pi.

Adds a fetch tool that retrieves a URL and returns the page content, truncated to 50KB / 2000 lines.

Fetches http(s) URLs.
Always asks for HTML (Accept: text/html, application/xhtml+xml, */*). Only some sites support returning text/markdown, and converting HTML generally does a better job cleaning up real-world pages.
Converts HTML pages to clean, LLM-friendly text.
If the response isn’t HTML, it returns textual content as-is; binary content is saved to a temp file.

This repo is a Pi package (see package.json#pi.extensions). Once installed, Pi auto-discovers and loads the extension.

Global install (writes to ~/.pi/agent/settings.json):

Project-local install (writes to .pi/settings.json in your project):

Try the package for a single run:

Requires curl installed and available in PATH.
Output is truncated to keep LLM context usage predictable.
The implementation is intentionally pragmatic and may look a bit ugly. It was largely generated with an LLM and optimized for “works reliably” over elegance.
The tool prioritizes efficiency and keeping output compact/predictable for LLM context, even if that means it won’t choose the best possible representation for every site.
Extensions execute with full system permissions. Only install code you trust.

GPL-2.0-only