pi-ketch-web-access

Unlimited web search for your AI agent. No API keys. No rate limits with SearXNG. No MCP server.

Four native tools — web_search, web_scrape, web_docs, web_code — that give your pi agent the web.

pi install npm:pi-ketch-web-access

---

Why Ketch? Because SearXNG

Ketch supports SearXNG as a search backend — a free, open-source metasearch engine aggregating 269 search services (Google, Bing, DuckDuckGo, Brave, Wikipedia, arXiv, and 264 more). Zero tracking. Zero profiling. No filter bubbles — unpersonalized results, not shaped by ad targeting. SearXNG also strips tracking parameters from result URLs.

Self-host it on your local machine, a $5/month VPS, or a Raspberry Pi running Docker. SearXNG gives you search across general, images, videos, news, maps, music, IT, science, and more.

• No API keys — no signups, no tokens to rotate, no billing
• No rate limits — unlimited programmatic access from your own instance. Upstream engines may throttle; results still arrive from the rest
• No filter bubble — unpersonalized results, not shaped by ad targeting

Ketch also supports Brave, DuckDuckGo, and Exa as alternative search backends (rate-limited; API keys may apply). Note: Brave is ketch's default backend — configure SearXNG to unlock unlimited self-hosted search.

---

Four tools. Zero friction

Tool	What it does	When to use it
web_search	Search the web. Set scrape: true to find and read pages in one call	General queries. Avoids the two-call round-trip that burns rate limits
web_scrape	URL → clean markdown. Auto-detects JS-rendered pages, re-fetches via headless Chrome (requires ketch browser install)	You already have a URL. Cached results stale? noCache: true
web_docs	Context7 version-aware documentation lookup (free tier; API key recommended for heavy use)	Library/framework API questions. Stops hallucination from stale training data
web_code	grep.app search across 1M+ open-source repos	Finding real implementation patterns. Language filtering

Every tool returns structured JSON. Every parameter is typed. Every tool ships with promptGuidelines that route the agent to the right tool without you writing a single line of prompt engineering.

---

Why an extension, not a skill?

Skills provide instructions; extensions register tools. This extension turns ketch CLI commands into native pi tools with typed parameters and structured output. The LLM calls them directly — no CLI syntax to fumble.

---

Install

Prerequisites:

• pi installed
• ketch CLI on $PATH
• (Optional) ketch browser install — enables JS-rendered page scraping for web_scrape
• (Optional) Install and configure SearXNG — unlocks unlimited self-hosted search

Install:

pi install npm:pi-ketch-web-access

---

Usage

No skill configuration. No prompt engineering. Install it and the four tools are available to your agent immediately.

"What's the latest Node.js LTS release date?"

→ Agent calls web_search

"How do I use the Zod v4 transform API?"

→ Agent calls web_docs — built-in routing tells it to prefer docs over search for library questions

"Show me a real implementation of a Rust async runtime"

→ Agent calls web_code with lang: "rust" — it knows code search beats web search for implementation examples

"Scrape https://example.com and extract the pricing table"

→ Agent calls web_scrape — JS-rendered? Headless Chrome handles it automatically

"Find and summarize the top 3 articles about MCP servers"

→ Agent calls web_search with scrape: true — find and read in one call, not two