Skip to content

v0.12.0

Choose a tag to compare

@nicobailon nicobailon released this 24 Jun 07:04

[0.12.0] - 2026-06-24

Added

  • Added webSearch.enabled config to disable web_search tool registration. Thanks @webwarrior-ws for PR #38 and @knocte for reporting #22.
  • Added workflow: "auto-summary" for web_search to generate a summary without opening the browser curator. Thanks @Ilm-Alan for PR #57 and @baflow for reporting #32.
  • Added Tavily Search API provider support with TAVILY_API_KEY / tavilyApiKey, domain filters, recency filters, and auto-provider fallback integration.
  • Added Parallel search provider support, auto-provider fallback integration, and Parallel fetch_content extraction fallback.
  • Resolved web-search.json from PI_CODING_AGENT_DIR, then XDG_CONFIG_HOME/pi, before falling back to ~/.pi.
  • Added GOOGLE_GEMINI_BASE_URL / geminiBaseUrl for routing Gemini generate-content calls through compatible gateways, plus CLOUDFLARE_API_KEY / cloudflareApiKey authentication for Cloudflare AI Gateway. Thanks @meatballhat-cf for PR #76 and @tynril for reporting #74.

Removed

  • Removed the deprecated code_search tool that duplicated the Exa search provider from web_search. Thanks @picasso250 for PR #62 and reporting #61.

Fixed

  • Kept curator sessions alive when browser auto-open fails and surfaced a copyable curator URL for Docker, WSL, and headless environments. Thanks @rca, @runningman84, and @k0valik for reporting #92, #93, and #55.
  • Updated @mozilla/readability to ^0.6.0 for GHSA-3p6v-hrg8-8qj7. Thanks @omar-elmountassir for reporting #86, and @henriquebastos and @av1155 for PRs #35 and #68.
  • Limited curator summary model selection to Pi enabledModels when configured, and fall back to a deterministic no-billing summary instead of silently calling unrelated catalog models. Thanks @Horace1423 for reporting #73.
  • Let configured Exa API keys use Exa's own account limits instead of blocking at the legacy local 1,000-request counter. Thanks @totoDoP for reporting #80.
  • Supported parallel web_search curator tool calls with per-call browser and cancellation state.
  • Prevented curator sessions from hanging after searches finish when no browser connects, and finalized connected idle sessions once the curator timeout elapses.
  • Added explicit role: "user" fields to Gemini generate-content requests for Vertex AI-backed proxy compatibility. Thanks @meatballhat-cf for PR #76 and @tynril for reporting #74.