@mmalecki/feather-pi-web

Lean web search and web fetch tools for pi, aimed at smaller models

Packages

Package details

extension

Install @mmalecki/feather-pi-web from npm and Pi will load the resources declared by the package manifest.

$ pi install npm:@mmalecki/feather-pi-web
Package
@mmalecki/feather-pi-web
Version
0.3.0
Published
Nov 18, 2026
Downloads
360/mo · 52/wk
Author
mmalecki
License
MIT
Types
extension
Size
23.5 KB
Dependencies
5 dependencies · 2 peers
Pi manifest JSON
{
  "extensions": [
    "./index.ts"
  ]
}

Security note

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

README

feather-pi-web

Lean web search and web fetch tools for pi, aimed at smaller models.

  • webfetch - fetches a page and reduces it to the markdown of its main content (up to 50k characters). With compact, a model summarizes it (with an optional focus) before it reaches the agent's context. With raw, the page source comes back as fetched instead (up to 150k characters, or a quarter of the model's context window if that is less), for when the markdown is missing something. Images come back as attachments, converted and scaled the way pi's read tool does it (following the images.autoResize setting)
  • websearch - searches through Parallel's Search API, which returns markdown excerpts per result

Installation

pi install git:github.com/mmalecki/feather-pi-web

Or try it for one run with pi -e /path/to/feather-pi-web.

Configuration

  • PARALLEL_API_KEY - enables websearch
  • WEBFETCH_COMPACT_MODEL - the provider/model that summarizes fetched pages, e.g. a small local one. Defaults to the session's model.
  • WEBFETCH_MAX_RAW_CHARS - the most characters of page source a raw fetch returns, whatever the model's context window. Defaults to 150000, or a quarter of the context window if that is less.

Summarizing runs without thinking, and its token usage counts towards the session's.