v0.2.4

Highlights

• Reduced Full PDF token usage by automatically trimming tail-end reference sections before context assembly
• Expanded tail trimming to also drop common end-matter blocks such as Acknowledgments and Appendix references when they appear near the end of indexed PDF text

Notes

• This release focuses on making Full PDF mode more token-efficient without changing Balanced mode behavior
• The packaged plugin asset attached to this release is paper-chat.xpi

v0.2.3

Highlights

• Added two PDF context modes: Balanced for summary plus relevant snippets, and Full PDF for single-document full-text inclusion with truncation warnings when needed
• Improved PDF extraction reliability by reading Zotero full-text cache files directly, retrying during indexing startup, and avoiding false claims that the PDF cannot be accessed
• Added live estimated input-token display in the status bar, including threshold warnings and over limit indication before sending

Notes

• This release focuses on PDF-context quality, token visibility, and single-document full-text workflows for chat/completions providers
• The packaged plugin asset attached to this release is paper-chat.xpi

v0.2.2

Highlights

• Default temperature is now 1 for new installs, preference fallback, and settings initialization
• Markdown rendering is more robust for academic answers with tables, inline formulas, and CJK-adjacent bold text
• Math rendering now prioritizes academic content while still skipping fenced code blocks and inline code spans
• Assistant message rendering now uses parsed DOM fragment insertion instead of direct innerHTML assignment to avoid Zotero runtime incompatibilities
• Added finer-grained render diagnostics to distinguish normalization, math preparation, Markdown rendering, and fallback stages

Notes

• This release specifically fixes the cases where Markdown replies could fall back to plain text or trigger Zotero instability during rich-content rendering
• The packaged plugin asset attached to this release is paper-chat.xpi

v0.2.1

Highlights

• Real streaming assistant output with SSE support when the model provider supports streaming
• Automatic fallback to the previous pseudo-streaming output when streaming is unsupported
• Fixed the post-response conversation persistence failure that could surface as “Unknown error, please try again later” after a reply finished rendering
• Safer conversation storage fallback with message compaction/truncation to avoid losing a full turn
• Improved rendering robustness with plain-text fallback when Markdown rendering fails

Notes

• This release specifically improves the error scenario reported when using google/gemini-3.1-flash-lite-preview, where content could appear during output but end with an error after completion
• Streaming remains provider-dependent and will gracefully fall back when the endpoint does not return SSE

v0.2.0

Highlights

• Multi-PDF context with add/remove actions, priority ordering, and per-conversation persistence
• Visual context badges for PDFs and selected text
• Selected-text context capture improvements with explicit toggle support
• Conversation management (new, switch, rename, delete)
• Typewriter-style assistant output animation
• Updated icon assets and README documentation

Notes

• Context budget is automatically controlled when multiple PDFs are selected
• Invalid PDF entries are auto-pruned during request building

PaperChat v0.1.0

PaperChat v0.1.0

First public release of PaperChat for Zotero 7/8.

Highlights

• Sidebar chat experience optimized for narrow Zotero reader panes
• OpenAI-compatible API configuration in Zotero Preferences
• Optional PDF context selection with search support
• Current-reading PDF prioritized as default selected document
• Markdown + LaTeX message rendering in chat replies
• Improved message spacing and selectable/copyable chat content

Notes

• Works with OpenAI-compatible /chat/completions endpoints
• PDF context depends on Zotero fulltext indexing availability

Included assets

• paper-chat.xpi (plugin package)
• Source code snapshot (auto-attached by GitHub release)