Version 0.1.3 — faster, sharper word lookup. Tap-to-look-up rebuilt • Bundled Japanese-to-Chinese dictionary (217 k entries) ships with the app. Tapping a Japanese word in 語 mode returns the reading and Chinese meaning instantly, no network call. • Every word you've looked up before is cached on-device, so repeat taps stay instant even for words outside the dictionary. • A small chip beside the translation shows "dict" or "llm" so you can tell which layer answered. • The refresh button now routes around the dictionary so you can always ask the LLM for a second opinion. Word boundaries that match reading • Word mode used to split every kanji into its own tap target — tapping ギブアップ would only ever select アップ. Now Japanese text is segmented into actual lexical words (ギブアップ, 今世, 嫁ちゃん, …) so taps land on the unit you mean. Auto-crop after analysis • Captured pages are automatically trimmed to the recognized text plus a small margin, so the overlay isn't dominated by empty margins. Toggle in Settings → Reading. Polish and fixes • First-launch default target language is derived from the device's locale instead of being hard-coded. • Settings now shows a live count of cached word lookups with a clear button when there's anything to clear. • Re-analyze no longer crashes on pages that mix scripts.

Version 0.1.2 — share photos straight into Yomibuddy, plus a more usable overlay. Share into Yomibuddy • Yomibuddy now appears in the iOS share sheet for any image — Photos, the screenshot preview, Safari "Save Image", Files. • After sharing, open Yomibuddy and the photo flows straight into the capture screen with crop/adjust ready to go. Overlay rework • Horizontal text now fills the screen left-to-right, with a vertical scroll for taller pages — no more letterboxed black bars on the sides. • Vertical text (tategaki) fills top-to-bottom, with horizontal scroll for wider pages. • Dragging the divider scales the image to keep the active axis full, so resizing the translation panel never wastes space. • Re-analyze and 文/語 toggle moved to a dedicated action strip on the divider, alongside new previous / next page buttons that mirror the library's newest-first order. Smarter crop and adjust • Crop and Adjust now auto-position the corners around the detected page edges instead of starting flush with the photo. • Works on book photos, screen captures, and comics — anywhere a page-shaped region stands out. Direction is automatic • Removed the "Default direction" setting. The OCR pipeline now decides whether a page is horizontal or vertical from its paragraph shape, and the overlay adapts accordingly.

Version 0.1.1 — focused on speed, polish, and giving advanced users more control. Faster everywhere • Word lookups (tap-to-define in word mode) now return in under a second on the bundled free service. • Page translations and chat replies are faster too. • A typing indicator animates while the AI is composing a reply, so you can tell something's happening even before the first character appears. LLM mode picker • Top of Settings now offers two paths: "Free bundled" (a built-in AI service we curate) or "Custom" (bring your own API keys). • In Custom mode you can pair different providers and models with each task: translation, word lookup, and chat. Faster/cheaper for high-frequency word lookups; a smarter one for chat. • Custom mode supports OpenAI, Anthropic Claude, Google Gemini, and DeepSeek out of the box, plus a "Custom…" field for any model id the provider supports. Library and capture • Adjustable library size — set the FIFO cap from 5 to 100 pages in Settings. • Library thumbnails now show the colored sentence outlines for analyzed pages at a glance, so you can tell processed pages from drafts. • Closing the post-capture screen now saves your photo as a draft. Tap the tile later to resume. • Cropper opens with the full image selected and translucent corner handles, so the original stays readable while you adjust. • Auto perspective correction is more conservative — it only fires when a clear document fills most of the frame, no more silently cropping multi-panel photos. Sentence + word detection • Better paragraph splitting handles line wraps and mixed quotes more robustly. • Tapping between words in word mode no longer accidentally selects the whole sentence behind them. • Repeatedly tapping the same word in word chat no longer spams duplicate prompts. Reading + chat polish • The default sentence-mode prompt is now "Word and grammar explanation" (localized) and auto-sends on first entry, matching the word-mode behavior. • A configurable cap on the very first AI reply keeps initial responses focused (sentence: source length × multiplier, with a 100-character floor; word: a fixed character count). • Source-language hint in word lookup prevents kanji from being read as Chinese when you're reading Japanese (and vice versa).