$ cat ~/map/pi-web-access.md

Pi Web Access

Web search, content extraction, and video understanding for the Pi agent.

extensions✓ Oct 16

Facts

repository: github.com/nicobailon/pi-web-access ↗
GitHub Stars: ★ 1.6K
Forks: 297
license: MIT
npm / mo: —
last push: Oct 16, 2026

install

pi install npm:pi-web-access

## TL;DR

Web search, content extraction, and video understanding for pi: 20+ search providers with smart fallback chains, zero-config Exa, Codex-subscription search reuse, YouTube/screen-recording Q&A, and GitHub repos cloned instead of scraped.

## Key points

$Zero config out of the box via Exa MCP without an API key; a signed-in Codex subscription lets OpenAI web search reuse that auth, and an active Kimi Code Plan (via /login kimi-coding) enables Kimi search without a separate key.
$Smart fallback chains per capability — search walks SearXNG, Codex-backed OpenAI, Exa, Brave, Parallel, Tavily, Firecrawl, Jina, Kagi, and more in order until one answers.
$Video understanding: point it at a YouTube video or a local screen recording and ask about what is on screen — transcripts, visual descriptions, and frame extraction at exact timestamps.
$GitHub URLs are cloned locally rather than scraped, so the agent reads real file contents and gets a local path to explore instead of rendered HTML.

## FAQ

Q Which providers work without any API key?

Exa via its MCP endpoint needs no key, DuckDuckGo is keyless, Codex-subscription OpenAI search reuses your pi login, and a self-hosted SearXNG endpoint gives fully private search.

Q How do I add a specific provider?

Add the provider's API key or endpoint (Brave, Tavily, Firecrawl, Kagi, Serper, Gemini, Perplexity and others are supported); remote hosted page fetchers additionally require the explicit fetchRouting.allowRemoteHostedProviders opt-in.

## References

## Related entries

Last updated: Aug 31, 2026