pi-web-surf

Internet research extension for the pi coding agent — powered by ketch CLI (v0.12+).

Tools

Tool	Description
internet_search	Web search (DuckDuckGo, Brave, SearXNG, Exa, Firecrawl, Keenable). Supports --scrape for full content.
internet_scrape	Fetch URLs → clean markdown. Auto-detects SPAs, handles PDFs, CSS selectors.
code_search	Grep real OSS source across public repos (Grep.app, Sourcegraph, GitHub Code Search).
library_docs	Curated, version-aware library/API docs via Context7.
web_crawl	BFS same-host crawl → markdown per page. For multi-page docs sites.

Install

Prerequisites

brew install 1broseidon/tap/ketch
# or: go install github.com/1broseidon/ketch@latest

As a pi package

Add to ~/.config/pi/agent/settings.json:

{
  "packages": [
    "git:github.com/yanralapdy/pi-web-surf"
  ]
}

As a local extension

ln -s ~/sites/js/node/pi-web-surf ~/.pi/agent/extensions/pi-web-surf

Configuration

ketch reads config from ~/Library/Application Support/ketch/config.json (macOS) or ~/.config/ketch/config.json (Linux):

ketch config set backend ddg          # DuckDuckGo (free, no key)
ketch config set brave_api_key <key>  # or use Brave (free key from brave.com/search/api)
ketch config set context7_api_key <k> # for library_docs (free from context7.com)

Usage

In pi:

/reload

# The model uses these tools automatically based on task:
# - "search for X" → internet_search
# - "read this URL" → internet_scrape  
# - "find code that uses X" → code_search
# - "show me the docs for X" → library_docs
# - "crawl this docs site" → web_crawl

Requires

• ketch >= v0.12.0
• pi >= 0.80.0

License

MIT