Pi Web Access Files

You can now search X posts through xAI or choose Mistral for web research. Multi-query searches run in parallel, so researching several angles takes less waiting. Follow-up retrieval can access stored search results, and Perplexity keeps the sources its answers cite. Proxy settings stay limited to web tools without affecting your other Pi requests.

Highlights

• Search X posts through xAI or choose Mistral for web searches.
• Get faster multi-query research with parallel batch searches.
• Retrieve stored search results and keep all cited Perplexity sources.
• Use a web-tool proxy without changing unrelated Pi requests.

Full changelog

Highlights

• Search X posts through xAI or choose Mistral for web searches.
• Batch searches run in parallel for faster multi-query research.
• Follow-up retrieval can access stored search results, and Perplexity keeps all cited sources.
• Web-tool proxy settings no longer affect unrelated Pi requests.

Added

• Documented the Linux xdg-utils dependency for automatic curator browser launch and the manual URL fallback. Thanks to @wickedTangent for issue [#336] and PR [#337].
• Added opt-in X post search through xAI alongside web search, configured with xaiSearchTools. Thanks to @Jerry2003sky for issue [#342].
• Added Mistral web search with provider: "mistral", using mistralApiKey / MISTRAL_API_KEY credentials. Premium web search is available as a separate opt-in. Thanks to @jaudiger for issue [#346].

Changed

• Run batch searches in parallel with a concurrency limit, keeping results in query order and preserving provider fallback. Thanks to @Noir-Lime for PR [#345].

Fixed

• Included responseId in search output so the model can retrieve stored results with get_search_content. Thanks to @axelbaumlisto for PR [#354].
• Fixed flaky tests on slower Node startups. Thanks to @axelbaumlisto for PR [#353].
• Kept all sources cited in Perplexity answers, even beyond numResults. Answers without citations still respect the result limit. Thanks to @schlessera for issue [#340] and PR [#341].
• Limited configured proxies to web-tool requests so unrelated Pi requests are unaffected. Thanks to @alexei-ciobanu for PR [#339] and @mystery4f for PR [#343].
• Cleaned up abandoned GitHub clone directories after crashes, only when their owning process is confirmed dead. Thanks to @yazanabuashour for issue [#331].
• Kept using an existing ~/.pi/web-search.json when XDG_CONFIG_HOME is set but no config file exists there. Thanks to @hu3rror for issue [#333].