v1.4.0

Changelog

All notable changes to this project will be documented in this file.

Full Changelog: v1.3.0...v1.4.0

[1.4.0] - 2026-05-09

Added

• Local AI Chat (Ollama integration) — new top-level Reader↔Chat toggle in the header. Streams responses from a local Ollama server (/api/chat), with model picker auto-populated from /api/tags and configurable host/port (default localhost:11434).
• Plain text (.txt) file support — drag, drop, or pick a .txt file to read it the same way as PDFs. Files are paginated into pseudo-pages of ~40 sentences and use the existing TTS, sentence highlight, library, and resume-progress flow. New TextPageRenderer component; fileType discriminator on library records.
• Multi-session chat memory — every chat is auto-saved to IndexedDB (new chat_sessions store, DB migrated to v3). Sessions list in the sidebar with switch / rename / delete; auto-named from the first prompt; LRU-capped at 50 sessions.
• Per-session event log — each session records sent / received / aborted / error events with timestamps. Viewable as a collapsible "Session log" section in the sidebar.
• Per-message read-aloud controls — every assistant bubble has a "Read aloud" button to start/stop TTS for that message specifically. Works with auto-TTS off; you can re-read finished messages later.
• Per-message copy button — one-click copy of any chat message (user or assistant) to clipboard, with toast confirmation.
• Markdown rendering in chat — assistant replies render as proper Markdown (lists, code blocks, tables, headings, links) via react-markdown + remark-gfm. User messages stay plain.
• Reasoning model support — toggle "Enable thinking" in the chat sidebar to send think: true to Ollama and surface message.thinking traces (for deepseek-r1, qwen3-thinking, gpt-oss, etc.). Thinking renders in a collapsible disclosure that auto-expands while streaming and auto-collapses once the answer arrives.
• Streaming vs after-complete TTS modes — toggle between reading sentences as they stream in, or waiting for the full reply. Mode is locked at message-start so toggling mid-stream applies to the next message.
• Markdown-aware TTS — chat speech strips Markdown markup (**, __, #, fenced code, link URLs) before synthesis, so the audio reads visible text only — no more "star star bold star star".
• Collapsible chat sidebar sections — Settings, Sessions, Conversation, and Session log are independently collapsible with their own scroll containers, plus an outer scroll for the whole sidebar.

Changed

• TTS engine refactor — extracted reusable synthesizeText, playChatUrl, playChatSpeech, and stopChatPlayback helpers; the chat side composes them on its own audio channel (chatAudioRef) so chat playback and reader playback can never collide. Reader playback is now gated by an enabled prop and pauses automatically when switching to chat mode.
• Bounded chat TTS prefetch — chat synthesis now mirrors the reader's pattern: only the current sentence plus the next two are in flight at any moment. Eliminates the cascading per-sentence timeouts that previously caused chat TTS to stop after a few sentences on long replies.
• Centralized toast + IndexedDB transactions — single transaction in saveBook/updateBookMeta, fewer round-trips, race-condition-free. Toast helper consolidated in App.jsx.

Fixed

• Stop button for chat TTS — clicking Stop on a chat message now silences audio immediately, drops queued sentences, and resets the playback chain (previously the in-flight sentence and queued ones kept playing).
• Long chat replies cutting off mid-message — the prefetch bound (above) prevents Kokoro's serialized inference from timing out queued sentences and silently skipping them. Replies after horizontal rules and other mid-message structure now read all the way through.

[1.3.0] - 2026-03-05

Added

• Health check endpoint (GET /v1/health) for frontend availability detection.
• Configurable API request timeouts and unlimited batch timeout toggle.
• Collapsible settings section in the sidebar.
• Duplicate synthesis request prevention (deduplicates in-flight fetches).
• Backend auto-failover with toast notification when Kokoro is unreachable.

Changed

• Major refactor: Frontend split into modular React hooks (usePdfEngine, useTtsEngine, usePersistedState, useKeyboardShortcuts, useMobileDetect, useTheme).
• Major refactor: Backend restructured from single server.py into a server/ package (app.py, endpoints.py, model.py, schemas.py).
• Entry point changed from python server.py to python run.py.
• Removed verbose TTS endpoint logging for cleaner server output.
• State management optimized — removed unused variables and refined useState/useEffect patterns.

Fixed

• TTS auto-continue on page change — reading now automatically continues on the next page without needing to pause and replay. Fixed race condition where the playback loop would re-run with stale text from the previous page.
• StrictMode double-run — playback index update reordered to prevent React StrictMode from skipping sentences.
• Improved useEffect dependency linting across hooks.

[1.2.0] - 2026-02-02

Added

• GPU support for Kokoro TTS (CUDA, OpenVINO for Intel Arc/NPU).
• "Continue from here" and "Selective Read" features in PDF Reader.
• Batch synthesis endpoint (/v1/batch_synthesize) for merging audio.

Fixed

• UI overlaps on mobile devices.
• PDF frame resizing and header layout issues.

[0.1.0] - 2026-01-31

Added

• Initial release