Pi Web Access Files

Pi Web Access 0.26.0 adds XCrawl as a new explicit search option and makes recent search and extraction paths less fragile. XCrawl results now keep source links usable even when the API returns redirect-style links. Local models that squeeze multiple searches into one JSON-looking query string now get separate searches instead of empty results. HTML fallback extraction is also quieter when a page cannot be processed cleanly.

Highlights:

• web_search now includes XCrawl as an explicit search provider.
• XCrawl results now produce safer clickable links, including relative redirect links from the API.
• Local models can send multiple web-search queries as a JSON string and still get separate searches.
• HTML fallback extraction is quieter and more accurate when Defuddle cannot process a page.

Full changelog:

Highlights

• web_search now includes XCrawl as an explicit search provider.
• XCrawl results now produce safer clickable links, including relative redirect links from the API.
• Local models can send multiple web-search queries as a JSON string and still get separate searches.
• HTML fallback extraction is quieter and more accurate when Defuddle cannot process a page.

Added

• Added XCrawl as an explicit-only search provider (provider: "xcrawl") with xcrawlApiKey / XCRAWL_API_KEY credentials, client-side domain filtering, fallback titles, and retriable provider-timeout errors. Thanks to @zeroicey for [#312] and [#313].

Changed

• Refined the XCrawl provider docs, parser shape, availability metadata, and focused tests.

Fixed

• Expanded JSON-array strings supplied in the singular web_search query field into independent searches, with a clear no-query error for empty arrays. Thanks to @alex-rs for [#317] and @zeroicey for PR [#319].
• Kept Defuddle selector-processing failures out of the Pi console and stopped treating the raw page body as a successful extraction. Thanks to @riabiy for [#315].
• Resolved API-origin-relative XCrawl result links to absolute URLs and rejected blank result links instead of emitting fabricated source URLs. Thanks to @zeroicey for [#318].