logancyang/obsidian-copilot 3.3.2 on GitHub

Release v3.3.2

This patch is all about stability and speed. Claude Opus 4.7+ now works correctly with adaptive thinking on both direct Anthropic and AWS Bedrock, project-mode file ingestion is fixed, Quick Ask no longer crashes on @-mention, and the plugin is 75 KB lighter thanks to a clean dependency deduplication.

🧠 Claude Opus 4.7+ adaptive thinking is fixed — Adding claude-opus-4-7 (or any Opus 4.7+) as a model and enabling reasoning used to return a 400 error because the plugin was sending the legacy thinking.type=enabled shape, which Opus 4.7+ no longer accepts. It now correctly sends thinking.type=adaptive and requests a summarized display so the thinking block appears in chat. Both the direct Anthropic and AWS Bedrock code paths are fixed. (@logancyang)
📂 Project-mode file ingestion is fixed — PDFs, images, audio files, and Office docs added to a project's source files were silently failing to upload because Obsidian's CORS-bypass path can't handle native FormData streams. The upload path now uses requestUrl with a manually-constructed multipart body, matching how the rest of the plugin talks to Brevilabs. (@zeroliu)
💬 Quick Ask no longer crashes on @-mention — Opening the Quick Ask panel and typing @ to mention a note was crashing with "useApp() called outside of AppContext.Provider". Fixed by wrapping the panel's React root in the same AppContext that the main chat view uses. (@zeroliu)
⚡ 75 KB smaller bundle — @langchain/community is dropped (Jina embeddings now run against a hand-rolled Obsidian-native client), and openai is bumped to v6 to eliminate a duplicate copy of the SDK that was sneaking in alongside @langchain/openai. The plugin is now around 3.3 MB. (@zeroliu)
🔴 Clearer error when a Bedrock model ID is missing its inference-profile prefix — If you configure a bare regional Bedrock model ID (e.g. anthropic.claude-sonnet-4-6 without a global., us., etc. prefix), the error used to surface as raw JSON. It now shows a plain-English message pointing you to Settings → Models and listing the four valid prefix forms. (@logancyang)
🧹 React state management cleanup — 56 setState-in-useEffect anti-patterns were replaced with idiomatic React patterns across 21 files. Side effects include: the token-count badge in chat now updates correctly after regeneration, chain/setting toggles in ChatInput no longer flash stale state, and the draggable quick-command modal no longer resets your resize when content grows. (@zeroliu)

More details in the changelog:

Improvements

#2460 chore(deps): remove unused deps and dead code to shrink bundle @zeroliu
#2463 chore(deps): drop @langchain/community + bump openai to v6 to dedupe SDKs @zeroliu
#2467 chore(react): centralize React root creation via createPluginRoot helper @zeroliu
#2469 Remove dead ChainFactory and chain validation helpers @zeroliu
#2470 style(css): expand #ccc to 6-digit hex format for consistency @zeroliu

Bug Fixes

#2399 fix(brevilabs): rewrite uploads via requestUrl multipart (W8/9) @zeroliu
#2454 chore(lint): fix no-direct-set-state-in-use-effect violations @zeroliu
#2466 fix(quick-ask): provide AppContext to overlay panel @zeroliu
#2471 fix(anthropic+bedrock): adaptive thinking + summarized display for claude-opus-4-7+ @logancyang
#2472 fix(bedrock): explain inference-profile requirement when AWS rejects on-demand model ID @logancyang

Generated by the release agent.