@tian.zuo/pi-web-search@0.7.0

Minor Changes

• #29 66de02a Thanks @TianZuo555! - Simplify /websearch-order keybindings: space now grabs/drops an item and enter always saves immediately, so reordering no longer requires a second confirm keypress. esc still cancels.

• #30 a42e967 Thanks @TianZuo555! - Add DeepSeek as a search provider using the official server-side web_search tool on DeepSeek's Responses API. Credentials are auto-detected from your pi DeepSeek login, then DEEPSEEK_API_KEY, then /websearch-auth. Agentic multi-round search with a synthesized answer; ranks after OpenAI in the search fallback chain; deepseek.model/deepseek.reasoning (default low) configurable in ~/.pi/web-search.json.

@tian.zuo/pi-find@0.4.0

Minor Changes

• #35 4c46ca1 Thanks @TianZuo555! - Harden grep/find against pathological searches: every rg/fd run now has a 30s wall-clock budget (SIGKILL on expiry, partial results kept with a "timed out, narrow the path" notice — never reported as a clean empty result), and rg skips files over 4MB during directory traversal so giant cache/bundle blobs can no longer turn a broad search into an overnight scan.

@tian.zuo/pi-antigravity@0.10.1

Patch Changes

• #36 c1184d7 Thanks @TianZuo555! - Fix print-mode string prompts and preserve consecutive user messages so extension-added context no longer replaces the caller's request. Use the same request boundary when restoring history and retain string-form historical messages.

@tian.zuo/pi-web-search@0.6.0

Minor Changes

• #27 35a256d Thanks @TianZuo555! - Expand /websearch-order into a tabbed editor for both fallback chains. The dialog opens on Search and switches to Fetch with Tab, preserves edits across tabs, and saves complete searchOrder and fetchOrder arrays together. The Fetch tab uses the same grab-and-move controls and includes the built-in direct provider.

@tian.zuo/pi-web-search@0.5.0

Minor Changes

• #24 5e0a1d1 Thanks @TianZuo555! - Add /websearch-order: an interactive grab-and-move dialog to reorder the search fallback chain (enter grab, ↑↓ move, enter save, esc cancel). Saves the complete searchOrder in ~/.pi/web-search.json; unconfigured providers keep their chosen position but are skipped until credentials become available. /web-search and /websearch-auth now point to it for reordering. The config file now lives at ~/.pi/web-search.json (the old ~/.config/pi-web-search/config.json is still read if present, and migrates on the next save).

  Also add openai.reasoning ("low" | "medium" | "high", or OPENAI_SEARCH_REASONING): sets the Responses API reasoning effort for search calls. By default the effort now follows the session's pi thinking level (mapped through the model registry); with neither set the model default (medium) applies. "low" is ~40% faster in practice and usually plenty for search.

Patch Changes

• #24 5e0a1d1 Thanks @TianZuo555! - Fix OpenAI/Codex detection being invisible: add a /web-search status command (provider credentials + active search/fetch chains), list openai as a read-only auto-detected row in /websearch-auth, and warn when the stored openai-codex token has expired (re-run /login) instead of silently skipping it.

• #24 5e0a1d1 Thanks @TianZuo555! - Attribute answer-only OpenAI searches to their internal source: when the Responses API answers from an internal tool (e.g. oai-weather, oai-finance) with no web URLs, the tool result now says answer via openai (internal source: oai-weather) and notes the internal source for the model, instead of showing a bare 0 results.

@tian.zuo/pi-antigravity@0.10.0

Minor Changes

• #22 be0df7c Thanks @TianZuo555! - Add gemini-3.8-flash (Gemini 3.8 Flash, low/medium/high efforts) to the fallback model catalog from the agy 1.1.25 agy models snapshot, with $0.75/$3.75 per-million-token reference pricing. gemini-3.5-flash, no longer served by current agy, is dropped from the fallback snapshot; live discovery continues to surface whatever the installed agy reports.

@tian.zuo/pi-background-terminals@0.5.2

Patch Changes

• #18 da8a245 Thanks @Alx8g! - Batch background terminals that finish close together into one bounded completion follow-up while preserving exactly-once delivery and the existing singleton message shape.

• #20 a45525c Thanks @Alx8g! - Preserve the full quiet window for arrivals that follow a busy-held expiry, keep the maximum hold fixed, and charge truncation markers to the strict aggregate completion budget.

@tian.zuo/pi-image-cache@0.1.2

Patch Changes

• #15 59a2c0e Thanks @TianZuo555! - Fix concurrent-session cleanup deleting freshly created cache directories by adding a 10-minute grace period, dispose the stale runtime on extension reload, clear the terminal display cache per session, evict display entries as a true LRU capped at 50, and separate model-facing placeholder strings into lib/prompt.ts.

@tian.zuo/pi-antigravity@0.9.0

Minor Changes

• #17 4f82756 Thanks @TianZuo555! - Require agy 1.1.22+, select the newest compatible installed stable binary with automatic cache invalidation, reuse a persistent stream-json driver across ordinary turns, report spawn/recycle counters and causes, recycle safely when model/profile/bridge configuration changes, add binary and model-effort diagnostics, expose custom agent and plan-mode configuration, enrich conversation status, and add safe direct markdown artifact previews.

Patch Changes

• #14 f1ae49c Thanks @TianZuo555! - Let agy own native context compaction through a 1M Pi scheduling window, persist safe same-session conversation resumes, render detected agy compaction boundaries, isolate Pi fallback summaries, report per-response usage, and show at most one substantive thought row per turn.

@tian.zuo/pi-usage@0.2.0

Minor Changes

• #12 d42073b Thanks @TianZuo555! - Support top 10 models in the /tokens dashboard with up/down arrow and j/k scrolling.