v0.16.0

Minor Changes

• f214f79: The library assistant now keeps its own conversations. Opening it no longer shows a conversation left over from a book you were just reading, and it remembers and restores your last library conversation when you come back to the library. Its empty state also reads sensibly outside a book.
• a53df57: The in-book AI assistant can now reach your whole library: ask it to compare the current book with others you've read, recall highlights and notes from any book, or check your reading stats — without leaving the book. It also gained a direct way to fetch the current book's whole-book summary.

Patch Changes

• abeae14: Fixed the in-book AI assistant reading and summarizing only a sliver of some EPUB chapters. When a chapter's text is split across several internal files (common in many EPUBs), the assistant previously saw just the first fragment — often only the chapter title and opening line — and missed the rest of the body. It now reads the chapter in full.
• eae95b2: AI memories are now purely global notes about you, no longer tied to the book they happened to be recorded in. The memory list in Settings no longer shows a source-book label, and memories are fully independent of your library — removing a book never touches them.

v0.15.0

Minor Changes

• 6d35b47: Give the AI companion awareness of the current date and time. Each turn now injects the wall-clock time (in your local timezone, ISO 8601 with offset) into the live message sent to the model, so it can answer time-relative questions ("how recent is this", "what year is it now") and reason about recency instead of guessing. The timestamp rides along only on the current turn — it never enters the cached system prefix and is never persisted, so it stays accurate every turn without disturbing prompt caching.
• c3fee3c: Added an optional background pass that periodically catches up on memories the assistant missed and tidies existing ones, with a low-key toast when it runs. Off by default — enable it under Settings → Memory.
• 43ade04: Add a library AI companion: a floating assistant in the library lets you chat with your AI assistant about your whole collection, not just one open book. It can browse your catalog, summaries, notes, highlights, and reading stats through tools — and, grounded in its memory of you, discuss your shelf and recommend what to read next. The same assistant panel is reused everywhere (including its conversation list and web-search toggle); conversations are no longer tied to a single book.
• fab63cf: Add an AI web search tool. Turn on the web search button in the AI panel's context row and the assistant can search the web (powered by Exa) for current or external information beyond the book — searches show up as a step in the reply and sources are linked inline. Works out of the box: Exa's free tier needs no API key. It's off by default; flip it on (the choice is remembered) and optionally add an Exa API key under Settings → Web search for higher rate limits. Backends are pluggable with automatic fallback so you can point it at another MCP search server.

Patch Changes

• d516a4c: Enable prompt caching for Anthropic (Claude) models, cutting cost and first-token latency on multi-turn conversations. The stable prefix that gets resent every turn — system prompt, tools, and prior messages — is now marked as cacheable (a fixed breakpoint on the system prompt plus rolling breakpoints on the last two turns), so repeated context is billed at the reduced cache-read rate and processed faster instead of re-charged in full each time. Caching is applied through a per-provider strategy layer: Claude (which has no implicit caching) gets explicit cache breakpoints, while OpenAI, Gemini, and OpenAI-compatible providers (e.g. DeepSeek) are passed through untouched since they cache long prefixes automatically server-side.
• 9c2e2e8: Fix PDF zoom interactions. Pinch and Ctrl+scroll now stay anchored to the cursor (and the toolbar +/− buttons and percentage input to the viewport center) instead of jumping to a different page — including during continuous pinch gestures, which previously drifted unpredictably. Zooming no longer flashes white: pages stay visible by stretching the current frame during the gesture and re-rendering to the new resolution once it settles, swapped in through an offscreen buffer so the canvas is never cleared on screen. Pinch/scroll sensitivity is also increased for a more responsive feel.

v0.14.0

Minor Changes

• b1dab8d: Add a customizable assistant avatar shown in conversation. Upload your own image (or keep the built-in default) in Settings → Agent; the avatar appears beside the assistant's messages along with its name, grouped so consecutive replies show it once. The avatar can be toggled off in settings.
• 8945d88: Crop the assistant avatar when uploading: pick an image, frame it with a round 1:1 crop (zoom + drag), and only the cropped result is saved.
• ca4c3eb: Recover from missing book files: when a book's stored copy is gone, the reader now shows a recovery panel offering to relink the original file (restoring reading progress and annotations) or remove the book from the library, instead of a dead error message.

Patch Changes

• b46de4d: Fix epub highlights showing no chapter in single-file books where every chapter shares one spine file (table of contents split by anchors). Annotation chapter lookup now subdivides such a shared file to the correct anchor chapter using boundary CFIs precomputed when the book opens. Multi-file books are unaffected.
• 75b3644: Add a lightweight update check: the app now checks GitHub Releases on startup and shows a toast when a newer version is available, plus a "Check for updates" button in Advanced settings. It only notifies and links to the release page (no auto-download/install yet).
• 5862712: Fix the selection toolbar's "Ask AI" action wiping out text you'd already typed in the chat composer. It now attaches the selected passage as context and keeps your draft intact, so you can finish the question you started. The preset actions (Explain / Translate / Summarize) still fill in their prompt as before.
• fd0cb40: Add a configurable global concurrency cap for background AI tasks (chapter/book summaries, conversation naming, and long-conversation compaction). Set it in Settings → Advanced (default 3); excess tasks queue and run as slots free up. Foreground chat replies are never throttled.
• c993045: Fix epub highlights showing the wrong chapter (or none) in the annotations sidebar. Chapter lookup matched a CFI's spine position against the TOC-based order index — two different numbering bases — so any book with a cover or front-matter page before its chapters was mislabeled. Annotations now resolve their chapter from the spine href, the same way current-chapter tracking already does.
• f096fab: Fix AI chat failing partway through a tool-calling turn when using an OpenAI Responses–API provider (notably third-party gateways) with a reasoning model. The request used to error with "Item … not found. Items are not persisted when store is set to false." Reasoning is now sent inline instead of as a server-side id reference, so tool-calling conversations complete on stateless endpoints. Also fixed the send-failure hint that always told you to configure an "Anthropic" API key regardless of the provider you'd set up — it now refers to your API key and model generically.

v0.13.0

Minor Changes

• 9101f4a: Add read-aloud (TTS) for ePub books: play from the current position with automatic chapter continuation, paragraph highlight that follows the speech, a floating control bar (pause/resume/stop/speed), and per-language voice selection with preview in Settings. Voices are matched to the content language of each paragraph, not the UI language.

Patch Changes

• 7f6b774: Reduce blank flashes when scrolling quickly through ePub books. Sections above and below the viewport now pre-mount and load further ahead, so their content is ready by the time it scrolls into view instead of briefly showing an empty placeholder.
• a33c870: Fix large position jumps when scrolling up in ePub books. Section iframes now keep their estimated height from the moment they mount (previously they collapsed to the browser default for a few frames while reloading, forcing the reader to compensate mid-scroll), unmeasured sections are estimated from chapter text length instead of a fixed placeholder, and sections above the viewport pre-mount earlier so height corrections happen off-screen.

v0.12.0

Minor Changes

• 72f7925: Add standalone book-level notes: write multiple Markdown notes per book from the reader sidebar's new Notes tab, or review them from the library via right-click → View notes — no text selection required.
• d2ee886: Book note editing now uses a syntax-highlighted Markdown editor (CodeMirror): headings render larger, bold/italic/code/quotes show their real styling as you type, markers are dimmed, and lists auto-continue on Enter.
• 768751f: Make the reader's sidebar and AI panel resizable: drag the inner edge of either pane to adjust its width. Widths are remembered across restarts, and the collapsed peek drawers follow the chosen width.

Patch Changes

• c357b76: Import EPUB books that are unpacked directories instead of zip files (for example, books exported from Apple Books, or epubs unzipped by Calibre/Sigil). These previously failed to import with an EISDIR error; they are now packed into a standard EPUB on import. Directories that aren't valid EPUBs, or whose contents can't be read (e.g. not-yet-downloaded iCloud placeholders), now report a clear, actionable error.

v0.11.0

Minor Changes

• ab3d830: Your reading companion now has a name (Lia by default), an editable persona, and a global long-term memory: it remembers your preferences, viewpoints, and recurring concepts across books and conversations, with full review/edit/delete control in Settings → Memory. Custom instructions replace the old assistant system prompt, and your existing chat model configuration is migrated automatically.
• 9cbd370: Add a gentle first-run onboarding card to the library that guides you to connect an AI model and turn on automatic chapter summaries. Skippable anytime, and it disappears once you're set up.
• bee788c: Seed a built-in localized sample book into the library on first launch, so new readers have something to open, annotate, and try AI on right away. It is a normal, deletable book and won't come back once removed.

Patch Changes

• 3594ad3: PDF reading polish: reopening a book now restores your exact in-page scroll position (not just the page), highlights with notes show a dotted bottom edge so you can spot them at a glance, and you can zoom smoothly with Ctrl+wheel or trackpad pinch — anchored at your cursor.

v0.10.0

Minor Changes

• 5eb8751: Add data backup & restore: export your whole library (books, annotations, reading progress, conversations, settings) to a single zip, and restore it on any machine. Restore replaces current data after a confirmation and keeps a pre-restore safety copy, then relaunches. Backups include your provider API keys in plaintext — the export dialog warns you to keep the file private.

Patch Changes

• 1b51661: Sort the annotations panel by creation time with the newest first, instead of by reading position — so freshly added highlights and notes always appear at the top.

v0.9.0

Minor Changes

• 9a3740e: Add reading-time tracking with a dedicated Stats tab: total / today / last-7-days, a daily reading chart, reading streaks, and per-book time ranking. Navigation now uses Apple Books-style top pill tabs (Library / Stats).

Patch Changes

• Show a pending placeholder in the AI panel while a reply is being submitted, so it's clear your request was sent before the assistant starts streaming.
• c0e2e48: Fix the AI assistant pretending to call reading tools in longer conversations. Past assistant turns now replay their real tool calls and results in history (instead of being flattened to text), so the model keeps actually reading the book instead of imitating a tool-free transcript.

v0.8.0

Minor Changes

• 4cbe610: Add a configurable step limit for AI replies (Advanced settings). The default agent step cap is raised from 5 to 10, and you can set any value from 1–99 or choose "Unlimited" — helpful for page-by-page PDF reading where the AI needs more tool-call steps to gather context.
• c0f3eca: Keep long AI conversations focused. Once a conversation's recent history grows past a large budget, older turns are folded in the background into a rolling summary while recent turns stay verbatim, so the prompt sent to the model stays bounded instead of growing without limit. This curbs the cost, latency, and quality drift (the assistant losing focus or skipping tool calls) that long reading sessions used to cause. Existing conversations are unaffected until they cross the threshold, and the summary is derived state only — your message history is never altered.

Patch Changes

• 3ac5f03: Auto-focus the AI panel input after triggering an AI action (ask/explain/translate/summarize, opening a conversation, or starting a new one), so you can type a follow-up immediately without an extra click.
• 77f49f5: Surface previously-silent failures in the AI streaming pipeline. When the main process couldn't push a stream chunk to the renderer (for example a payload that fails structured clone), the error was swallowed with nothing written to the logs; the chat stream's client-side errors were likewise never recorded. Both paths now emit warnings, and each agent step's finish reason is logged in development. So when an AI reply stalls or a requested tool call doesn't fire, there's a diagnostic trail to follow instead of silence.
• f73fe67: Mark books as finished in your library. Right-click a book and choose "Mark as finished" to flag it as read — a green check badge appears on its cover, and the book drops off the "Continue reading" shelf. Choose "Unmark as finished" to undo. The book's right-click menu now also shows an icon beside each action.
• fd08446: Edit, resend, and regenerate AI chat messages. Hover a message in the AI panel to reveal actions: edit one of your earlier questions and resend it, resend it unchanged, or regenerate the assistant's reply. The conversation is truncated from that point and a fresh reply is streamed — useful for rephrasing, retrying a failed reply, or getting an alternative answer.
• 837f323: Fix the provider connection test failing for reasoning models (such as Kimi K2.6) with a baffling "failed: HTTP 200" message. The connectivity probe only requested a single output token — far too little for a reasoning model to even begin its internal reasoning — so the provider returned an incomplete HTTP 200 response that surfaced as a self-contradictory failure (a success status code reported as an error). The probe now allows enough output budget for reasoning models, a 2xx response that still can't be parsed now reports an honest diagnostic instead of a bare status code, and the test failure — previously swallowed with nothing written to the logs — is now recorded as a warning.
• 1fecb84: Add a copy button to AI chat messages. Hovering (or keyboard-focusing) any message in the AI panel reveals a small toolbar; the copy action places the message's markdown source on the clipboard, so you can paste a reply into notes or feed it back into another prompt. Works for both AI replies and your own messages.

v0.7.0

Minor Changes

• 8e8fdbf: ePub reader now navigates at anchor granularity: clicking a chapter or an in-text link scrolls to the exact #fragment anchor, external links open in the system browser, and current-chapter highlight follows the anchor you're reading as you scroll. Reading position is restored to the exact paragraph you left off at — even for books that pack the whole text into one or two large HTML files, where it previously snapped back to the beginning. Also fixes a white screen when clicking in-text hyperlinks (the sandboxed iframe used to navigate itself to an invalid URL).
• 7b11735: ePub reader now recognizes chapters anchored within shared spine files (e.g. Calibre/Epubor exports that pack a whole book into one or two HTML files). The table of contents, AI chapter tools, and chapter text all resolve at #fragment-anchor granularity instead of collapsing to per-file chapters. Existing books upgrade automatically on next open.