What's new in v3.6.0: • Choose Cloud AI, prepare on-device AI for offline use, or decide later during onboarding. • See the AI type and model used for each answer in Chat. • Answers now render headings, lists, emphasis, code, and other Markdown formatting. • Improved retrieval order helps on-device AI keep the most relevant clauses in its limited context.

What's new in v3.5.4: • Fixed the on-device AI sometimes saying "the documents don't contain the answer" when the answer was in fact right there. In our testing this happened to about 1 in 5 answerable questions in English; it no longer happens, and the AI still says so plainly when the answer genuinely isn't in your documents.

What's new in v3.5.3: • The plan screen now explains the benefit that matches what you were doing: keep asking after the free limit, or unlock stronger Claude models.

What's new in v3.5.2: • The on-device AI now says plainly when your documents don't contain the answer, instead of stretching an unrelated passage to fit. In our testing this went from 58% to 96% of such questions. • Fixed answers starting in English when you asked in German, Spanish, French or Portuguese. Answers now begin in the language you asked in. • Fixed the on-device AI option being untappable before the model was downloaded. Tapping it now starts the download and shows the progress.

What's new in v3.5.1: • Fixed a problem where devices that cannot run the on-device AI could still download a model that would never work. Those devices now clearly say so instead. • Cloud AI continues to work on every supported device.

What's new in v3.5: • Fixed a problem where a failed import could look like it succeeded, even though nothing was actually brought in. • Fixed a problem where re-importing a document with the same name could delete the copy you already had, if the new import failed. • Switching to the on-device AI now starts downloading the model. Before, it only changed the setting, and you found out it was missing when you asked a question. • Fixed search for languages that separate words with spaces, such as English and German. • A question blocked by the daily limit now continues after you subscribe, instead of being lost. • Replaced the built-in sample document with something closer to a real one. You can try finding a single line in a long contract, reading a page that is only an image, and asking about a foreign-language clause in your own language.

Changes in v3.4: • Scanned documents are now recognised in your device's language, instead of only Japanese and English. • Text files saved in encodings other than UTF-8 — such as Windows-1252 and ISO-8859-1 — now open correctly. • Fixed headings and tables coming out scrambled in PDFs that mix Latin and Asian scripts. • Answers now retry automatically when the service is briefly busy, and error messages are easier to understand. • Importing long scanned documents is more reliable.

Changes in v3.3.2: • A maintenance update, with no changes to how the app looks or works. • Internal improvements that help us find and fix problems faster.

Changes in v3.3.1: • A maintenance update, with no changes to how the app looks or works. • Internal improvements that help us find and fix problems faster.

What's New in v3.3: • On first launch, 3 quick screens explain the basics and a sample document (PDF) is added automatically — just tap a suggested question to try it right away. • Fixed an issue where characters could come out in the wrong order in PDFs that mix Japanese and Western text, improving answer accuracy. • Added a button to the chat screen to add your first document in one tap. • Adjusted when the app asks you to rate it. • Fixed an issue where the question you typed was lost when you reached the free plan limit.

What's New in v3.2: • Cloud AI upgraded to Claude 5 — Claude Sonnet 5 is now the default (Standard & Pro), and Pro can also choose Claude Opus 5 for deeper reasoning and more accurate answers. • Previous models (Opus 4.8 / Sonnet 4.6) remain selectable. • Stability improvements.

What's New in v3.2: • Cloud AI upgraded to Claude 5 — Claude Sonnet 5 is now the default (Standard & Pro), and Pro can also choose Claude Opus 5 for deeper reasoning and more accurate answers. • Previous models (Opus 4.8 / Sonnet 4.6) remain selectable. • Stability improvements.

What's New in v3.1: • Cloud AI upgraded to the latest Claude models — Claude Opus 4.8 (Pro) and Claude Sonnet 4.6 — for smarter, faster answers. • On-device AI models upgraded to a higher-quality build, for better local answers at a similar download size. • Stability improvements.

What's New in v3.0: • Audio & Video Transcription: Import MP3, M4A, WAV, AAC, FLAC audio files and MP4, MOV, M4V video files. WhisperKit (iOS) and whisper.cpp (Android) transcribe everything entirely on your device. • AI-Powered Cleanup: An LLM post-processing pass automatically removes filler words and stutters for cleaner, more readable transcripts. • Tap-to-Timestamp: Tap any answer to jump straight to the timestamped segment in the original audio. • Now supports 23 file formats across documents, images, audio, and video. • Bug fixes and performance improvements.

What's New in v2.2: • Collection Sort: Sort collections by Name (A→Z / Z→A) or Created Date (oldest / newest) • Custom Order: Drag collections into any order you like • Bug fixes and performance improvements

Updated App Store preview videos and metadata.

Updated screenshots.

What's New in v2.1: • Inline Citations: AI answers now show [1][2] source references — tap to jump to the exact page • PDF Citation Highlight: Jump to cited page with yellow highlight on source text • Cross-Language Search: On-device embedding model (E5 Multilingual) for semantic search across languages • Search Pipeline Trace: See how your query was processed (collapsible trace below answers) • BM25 Search: Upgraded from TF-IDF to Okapi BM25 for more accurate retrieval • OCR Editor: Edit extracted text from images and re-index • Layout-Aware PDF Tables: Better extraction of tables from PDFs • Localized Recommended badge for On-Device models • Bug fixes and performance improvements

What's New in v2.0: • Local LLM: Run AI on your device — no internet required, complete privacy • AI Mode Selection: Switch between local LLM (offline privacy) and Claude API (maximum accuracy) • Improved source citations for cross-language queries • Performance optimizations and bug fixes

What's New in v1.3: • Excel/PowerPoint: View and search XLSX and PPTX files • Cross-Language Search: Search documents across languages — ask in English, find answers in Japanese documents • Improved RAG: Larger context window, better source citations • Now supporting 15 document formats • Bug fixes and improvements

What's New in v1.2: • Image Support: Import JPG, PNG, HEIC, and WebP files. On-device OCR extracts text for AI Q&A — now supporting 13 formats • Photo Library Import: Import images directly from your photo library • Collections (Folders): Organize documents in hierarchical collections (up to 3 levels). Search includes all sub-collection documents • Multi-File Import: Select and import multiple files at once with per-file progress • Image Viewer: View imported images with pinch-to-zoom • Bug fixes and performance improvements

New Features & Improvements - Enhanced chat copy: long-press the copy button to choose from 3 modes — answer only, with question, or entire conversation - Grouped citations: multiple pages from the same document are now displayed together (e.g., Document.pdf (p.5, 12)) - Shared text now includes citation information - Added iCloud sync toggle in Settings - Improved scroll position so answers are visible from the top