v0.6.6

v0.6.6

UI

• Fixed version display — version.json imported at build time, loading overlay shows real version instead of 0.0.0
• Simplified loading labels — "Starting Python backend..." → "Initializing...", "Loading translation memory..." → "Loading..."

Infrastructure

• .gitignore — added logs/ and *.novelproj
• bump script — fixed changelog insertion positioning and bumped from X to X reporting the same version

v0.6.5

• CRITICAL: Fixed missing import asyncio in context_gatherer.py — single-segment /api/translate was crashing with NameError
• CRITICAL: Fixed batch_answer_and_classify missing session.commit() — all QA answers, character/gender changes silently lost
• CRITICAL: Fixed loadSegments abort using stale activeChapterId — rapid chapter switching no longer shows wrong segments
• CRITICAL: Fixed ChapterNav rename using non-existent fetchChapters named export — rename now refreshes the list via store
• CRITICAL: Fixed IPC path ignoring AbortSignal — timeouts now work correctly in Electron builds via _timeout parameter
• Fixed case-sensitive == character/term lookups (changed to .ilike()) eliminating duplicate entries
• Fixed instructions/actions saved in separate transactions — now single transaction, consistent on failure
• Fixed Pydantic str fields without defaults on CharacterResponse and GlossaryResponse — no more 500 on NULL from DB
• Fixed str(e) leaking API keys/paths to client — added sanitize_error() with regex redaction + truncation
• Fixed multiple internal commits in _apply_answer_to_db — consistent transaction boundaries, callers manage commits
• Fixed answer_question missing session.commit() (was relying on _apply_answer_to_db's removed internal commit)
• Fixed novel store reference in useEffect dependency array — caused clear() + state loss on every unrelated store update
• Fixed handleReapply lacking concurrent translation guard — added if (translatingChapter) return
• Fixed missing chapter_id field in translation result segments (violated frontend Segment interface)
• Fixed apply_qa opening 1 DB session per segment — now uses single session for entire batch loop
• Fixed cancellation returning segments list shorter than total — remaining None entries filled as untouched
• Fixed cancelTranslation leaking AbortController in _activeControllers map
• Fixed ContextPanel context fetch race on rapid chapter switching — request counter discards stale responses
• Fixed consistency check errors silently showing "No issues found" — added error state with distinct UI
• Added user-visible error feedback for export failures

Changelog: See CHANGELOG.md

v0.5.0 — First Public Release

v0.5.0 — First Public Release

An AI-assisted webnovel translation workstation with persistent narrative context and proactive ambiguity resolution.

Features:

• Batch chapter translation (configurable 1–10 segments per LLM call)
• 15+ LLM providers: Ollama, OpenAI, Anthropic, DeepSeek, Google Gemini, Groq, Mistral, Together AI, Perplexity, Cohere, OpenCode, xAI, OpenRouter, Fireworks, Deep Infra, custom
• Sliding window context (3 previous + 3 upcoming segments)
• Persistent character management with auto-detect on import (Chinese/Japanese/Korean), gender suggestion from name
• Glossary auto-population from detected cultivation/cultural terms
• QA Queue: flags pronouns, unknown names, idioms, cultural terms without blocking translation; batch-resolve by type
• Per-project translator instructions with "View prompt" preview
• Transliteration: pinyin with accent marks (nǐ hǎo), pykakasi romaji, hangul romanization
• Live progress tracking with cancel button
• Resume support — skips already-translated segments
• Re-apply — clear and re-translate with updated context
• .novelproj cache — full project export/import with auto-save
• Export: plain text, HTML (escaped), markdown

Changelog: See CHANGELOG.md