KOAssistant v0.19.0 Release Notes

Settings reset recommended. After updating, go to Settings → Backup & Reset
and reset Quick Actions panel, Quick Settings panel, and Input Dialog actions to pick up new defaults.
Or do a full Reset. Custom actions, API keys, notebooks, chats, and cached artifacts are not affected.

Section Artifacts & Source Selection

Artifacts are no longer limited to the full document. You can now generate X-Rays, summaries, analyses, and other artifacts for specific chapters or sections.

• Section picker: Choose any chapter or section from the book's table of contents.
• Source selection popup: Before running, choose scope (full document or section) and source (full text, cached summary, or AI knowledge only). Remembers your last choice per action.
• Section summaries: Generate section-specific summaries on the fly — cached and reusable by other actions.
• Smart X-Ray lookup: Automatically selects the most relevant X-Ray for your current page (section > main > prompt to choose).
• Cross-section search: "Look up in X-Ray" searches across all section and main X-Rays simultaneously.
• Artifact unification: All artifact actions now use the same View/Update/Regenerate workflow. Previously, some cached silently in the background.
• → Chat button: Available in all artifact viewers — ask follow-up questions with the artifact as context.

Applies to: X-Ray, Recap, Summary, Analysis, Analyze Notes, Key Arguments, Discussion Questions, Quiz, Extract Insights, X-Ray Simple, and Reading Guide.

Pinned Artifacts & Starred Chats

• Pin any AI response as a named artifact (hold → "Pin to Artifacts"). Appears in Artifacts browser and popups.
• Star conversations to mark important chats. Starred chats appear in a virtual "Starred" folder in Chat History.

Research Mode

Automatic academic paper detection and adaptation.

• DOI detection from document metadata and first-page text.
• Academic prompt tracks: X-Ray uses research categories (Key Concepts, Methodology, Findings, Referenced Works), About becomes a structured overview, Find Similar focuses on related literature.
• Auto web search: DOI-detected documents enable web search for relevant actions.

Notebook Vault Storage

Notebooks can now be stored in a central folder or a custom location (e.g., an Obsidian vault), not just alongside the book.

• Three save locations: alongside book (default), central folder, or custom path.
• Vault/central modes use Author — Title.md filenames with YAML frontmatter.
• Switching location auto-migrates existing notebooks with collision handling.
• Notebook button now opens a popup: Add to Notebook / View / Edit.

Multi-Book Action Launcher

Multi-book actions (Compare, Find Common Themes, Analyze Collection, Quick Summaries, Reading Order, Recommend) now use a dedicated book picker with search and filtering, instead of requiring navigation to each book. New action: Recommend Books.

Action Changes

• AI Wiki: Encyclopedia-style entries from highlights or from the X-Ray browser (per-item, cached as artifacts).
• Reading Guide: Spoiler-free guidance for what to watch for as you read. Adapts to genre.
• Grammar: Sentence-level grammatical breakdown — word-by-word with optional constituency parse. Language-aware.
• Smart actions removed: Explain/Analyze in Context Smart replaced by the source selection popup on the standard actions.
• Action names shortened: "Book Info" → "About", "Analyze My Notes" → "Analyze Notes", etc.
• In-context trio clarified: Explain (comprehension), Analyze (reader-focused), Thematic Connection (craft analysis) — less overlap.
• Deep Analysis: Restructured for dense, scannable output.

New Providers

Z.AI

17th built-in provider. GLM models (GLM-4.5-Flash free tier) with regional endpoints (international/China) and toggleable thinking for GLM-4.5+ models.

Perplexity

18th built-in provider. Sonar models with always-on web search, citations as clickable footnotes, and reasoning extraction for sonar-reasoning-pro.

Streaming <think> tag extraction (also benefits Groq, Together, Fireworks, SambaNova, and Ollama).

Local Provider Presets

Quick setup for LM Studio, llama.cpp, Jan, vLLM, KoboldCpp, LocalAI. Pre-fills name and URL, no API key required.

Reasoning & Model Updates

Reasoning Expansion

New effort-level controls for always-on reasoning providers:

Gemini 2.5: Thinking now controlled via master toggle with configurable budget (dynamic/low/medium/high/max). Previously always-on with no user control.

Mistral: Magistral models now always think — reasoning extracted automatically.

Model Updates

• OpenAI: GPT-5.4 (new default) and GPT-5.4 Pro added. Reasoning is gated (controlled by master toggle, same as GPT-5.1/5.2).
• Gemini: Gemini 3.1 Pro Preview replaces 3 Pro Preview. Gemini 3 Flash Preview (free tier) and 3.1 Flash Lite Preview added. Gemini 2.0 models removed.
• Z.AI: GLM-5 (flagship) and GLM-4.7 family added.

Other Improvements

• Text selection popup in all KOAssistant viewers: 1 word → dictionary, 2+ words → Copy/Dictionary/Translate/Notebook popup.
• Recap reminder: Optional prompt to run AI Recap when opening a book after configurable days of inactivity.
• KOReader storage modes: All metadata modes now fully supported (previously only default mode was reliable).
• Chat viewer: Button layout streamlined. Dictionary view mode for highlight actions. Export saves to file (Copy handles clipboard).
• API errors: Now shown as readable text instead of raw JSON.
• Quiz: Answer Key moved to separate section to prevent spoilers.
• Kobo: No longer crashes without Wi-Fi at startup.

Bug Fixes

X-Ray & Artifacts

• Fixed cross-book browser/viewer showing wrong book's data.
• Fixed section X-Ray surfacing, scope matching, and popup labels.
• Fixed AI Wiki placeholders, detail refresh, and cross-navigation.
• Fixed X-Ray lookup skipping description matches.
• Fixed TOC ghost entries in section artifact pickers.
• Fixed XPointer hidden flows in page display.

Chat & Conversations

• Fixed chat titles, section scope labels, and starred status persistence.
• Fixed v2 chat delete and navigation gaps.
• Fixed group popups layering and multi-book context routing.

Providers & Reasoning

• Fixed Perplexity citation formatting and message merging.
• Fixed default thinking preservation for Gemini 2.5, DeepSeek, and Z.AI when master toggle is off.
• Fixed Anthropic 4.6 adaptive thinking not suppressed when reasoning config is "off".

Other

• Fixed UTF-8 context truncation for non-ASCII text (Arabic, CJK).
• Fixed notebook vault migration, collision handling, and export paths.
• Fixed action flag preservation when duplicating actions.
• Fixed "Keep English" not saving for non-English users.
• Fixed markdown link formatting in AI responses.

For Custom Action Users

• Smart actions removed: Recreate using standard versions with source_selection enabled.
• "Use Primary" → "Follow Primary": Language picker wording standardized.
• The source_selection flag is available for custom actions — add it to any text-extraction action.

Full Changelog: v0.18.2...v0.19.0