The best Pi web search extension?
I run the Pi coding agent with pi-web-access installed, and recently I compared it against pi-web-agent, the other popular web search package for Pi. They solve the same problem in two different ways.
pi-web-access is a toolbox. It gives the model direct tools like web_search, source_check, and fetch_content, and the model chains them itself. But pi-web-agent is a bounded research workflow. It exposes one tool, web_explore, which plans queries, picks candidate pages, reads them, and synthesizes an answer with caveats, all internally.
| pi-web-access | pi-web-agent | |
|---|---|---|
| Model interface | Direct tools, chained by the model | One web_explore call |
| Search providers | 20+, including zero-config Exa MCP | 6 backends, DuckDuckGo by default |
| Page reading | HTTP fetch plus readability extraction | HTTP plus headless Chromium rendering |
| Verification | source_check with passage citations |
Built-in evidence quality and caveats |
| Special content | GitHub cloning, PDFs, YouTube, video | None |
| License | MIT | AGPL-3.0-only |
For my setup, pi-web-access is the better default. It searches through Exa's MCP endpoint with no API key, while pi-web-agent's hosted backends need keys and otherwise fall back to scraping DuckDuckGo. pi-web-access also has source_check for verifying claims, plus GitHub, PDF, and video support that the other package does not have.
pi-web-agent wins in two places. It can render pages in headless Chromium, which handles bot checks and JavaScript-only sites that plain HTTP reads miss. And it is honest about weak evidence, reporting unreadable threads and narrow source sets as caveats instead of pretending. It also keeps transcripts much shorter in compact mode.
For everyday coding work, fetch this page, check this claim, search this API, pi-web-access covers more ground. For deep multi-source research where you want a disciplined workflow, pi-web-agent is the better fit.