Skip to content

README.md

Latest commit

 

History

History
75 lines (46 loc) · 2 KB

README.md

File metadata and controls

75 lines (46 loc) · 2 KB

pi web extension

npm version license

A pi extension that adds web search and web fetch tools to the coding agent.

  • Keyless web search via Brave and DuckDuckGo HTML fallback
  • Webpage fetching with HTML-to-Markdown conversion, saved to a temp file
  • Agent reads fetched content in chunks via the read tool — no context bloat
  • Automatic prompt steering for URL and web-search style prompts
  • Token-aware: keeps search results compact, trims oversized pages

Install

pi install npm:pi-web-extension
Alternative install methods

From the public git repo:

pi install git:github.com/NicoAvanzDev/pi-web-extension

From a local clone:

pi install .

Load without installing:

pi --no-extensions -e ./index.ts

How it works

websearch

Runs a keyless web search by scraping public search engines. Tries Brave first, falls back to DuckDuckGo HTML. Returns a compact list of results (title, URL, snippet).

webfetch

Fetches a URL, strips non-content elements, converts the HTML to Markdown via Turndown, and saves the result to a temp file in the pi session directory. Returns metadata (file path, title, content length, preview) so the agent can read the file in chunks as needed.

Prompt steering

Before each agent turn, the extension checks the user prompt for URLs and web-search intent patterns. When detected, it activates the web tools and injects steering instructions into the system prompt.

Tools

The extension exposes LLM-callable tools:

  • websearch
  • webfetch

websearch

Parameters:

  • query: string -- the search query

webfetch

Parameters:

  • url: string -- the URL to fetch
  • format?: "markdown" | "text" | "html" -- output format (default: "markdown")