Edit the PowerPoint you already have Every slide, as soon as it opens • A .pptx opens with a thumbnail of every slide, and each slide's text boxes, pictures, shapes and groups listed out. Pictures show a thumbnail, so you can tell which one is which. • Edit text paragraph by paragraph: size, bold, italic, underline, colour, alignment — the slide preview follows as you adjust. • Type a position or size, or use the steppers, and see it straight away. Pictures keep their proportions by default. • Add text boxes, shapes and pictures, change what sits in front, delete what you don't need. Slides can be added, duplicated, deleted and reordered. • Pictures inside a group can be replaced and adjusted one by one. Change a deck by asking • Tell the AI what to change — wording, colours, text colour, a picture, a new shape, what's in front. The approval card shows the slides before and after; nothing is written until you agree. • Models that can see images check the result first and fix it themselves before handing it to you. • Change a picture with AI, or replace it with a new one: the original and the new one side by side, applied only if you like it. • Leave a note for the AI on any slide, and apply them all at once. Slides that are only a picture can still be read • On-device text recognition is free and never leaves the device. Hard slides can be read by the AI one at a time. • Review and correct each slide, then save the lot as a Markdown outline — ready to rebuild into a new deck with the export wizard. File versions in the chat • Every approved change leaves a version card in the chat: look back, compare with now, restore, or branch a new chat from that version. • Exported decks appear in the chat as a thumbnail card. Tap it to return to the wizard as you left it. Fixes and improvements • Scrolling a deck, Word, Excel or PDF file — or an edit sheet — on iPhone no longer swipes it closed. • Documents open at nearly the full screen on Mac and iPad. • Esc cancels an edit sheet on Mac. • On an iPhone turned sideways, the edit sheet's preview no longer pushes the buttons off screen. • Files the app creates — exports, files the AI writes — show up in the file list straight away.

Slide export, one page at a time A four-step wizard: source → imagery → outline → output • Exporting a deck is no longer press-and-pray. Each step does one thing, and each one shows you its own result. • Step one takes reference material: knowledge base entries, files from your working folder, or a URL — pasted links are fetched as a snapshot and kept in the knowledge base. • The quote appears while you are choosing the imagery, and it counts images rather than pages. Anything already generated is not billed again. See the whole deck before you pay • The preview is what you will get: every page's layout, its words, and the prompt that page will use to draw its picture. • Edit any page's text. You can also direct just one page's picture ("watercolour for this one") — that note never gets printed onto the slide. • Not quite right? Change one part of that image and keep the rest. • Want to check the style first? Generate a single image. That one is the final one: the export reuses it instead of billing you twice. • A set of edits can be thrown away all at once, back to the version you had. • Changed pages are marked, and only they are redrawn. Every other picture is reused. AI Refine now really follows your settings • The refine settings moved next to the deck they shape, so one tap redoes it — no trip back to Settings. • Style now changes how pages are built, not just their tone. "Minimal" really does land on 5–7 pages. • The outline is written in the document's own language, and it says which language that is. • Auto-illustration now lets the model mark the pages that deserve a picture in the outline itself, so those pictures can be previewed, edited per page, and tested one at a time. • Full-page images no longer stop at 12 pages. A long deck simply costs more, and the quote says so before you start. • A failed refine now says so. It used to look exactly like "nothing changed". • Knowledge base reference material now reaches all four imagery modes. Fixes • Exporting a PDF on a Mac could never finish. • The checkmarks in menus all disappeared, so you could not see which tools were on. • Three academic search backends had been broken. • Needing a subscription used to fail first and explain afterwards. It now opens the subscription page. • A failed export reports the real reason instead of guessing at a settings page for you to go and change. Also • Two new web tools: turn one page into Markdown, and list a whole site's URLs.

Previews your way, and everything you tune can travel Reading appearance - Set the font, size, line spacing, reading width and colours of the Markdown preview, with their own styles for quotes, lists, tables, code and callouts. Five built-in looks change the preview the moment you pick one. - Adjust while you watch: what you see is the real preview. It only affects reading; exported PDF and Word files don't change by a single character. Bigger, easier-to-read text - Text across the app moves up one size, and the few small labels that ignored your system text size now follow it too. Sharing: all twelve kinds - Brand kits, the four export templates, reading appearances, tasks, scripts and provider setups now export as files — twelve kinds with prompts, export styles and KB entries. - Each one brings what it needs: a template carries its brand kit, a task its prompts and KB entries, a script the files it reads. - A provider setup carries the connection only: the API key and custom header values are never written into the file. Safer - Before the AI sends a request that could change data at an external API (such as a POST), you see a card: where it goes, exactly what it sends, and whether your key goes with it. Nothing is sent until you allow it. Also new - Export a conversation as Markdown into your workspace. - Smarter KB search: generate index cards so a question finds an entry even when it uses different words. - Changes made with the formatting bar can be undone (⌘Z). - Add and edit templates right from the export panel; "Style" lists only the fields this export will use. PDF merge now has a way in. - Two more models for AI illustrations. Fixes - Some newer models failed with an error. - On Mac, items in the brand kit and export template lists couldn't be deleted. - Some labels still showed Traditional Chinese in English and Simplified Chinese. - Browsing a source without an index file wrongly said "needs a newer app". - Importing a style pack now names the items it won't import instead of skipping them silently. - Sharing an item whose title had a line break could lose part of its details.

Formatting without remembering Markdown A formatting bar - Select some words and tap once for bold, italic, inline code or strikethrough. Tap again to remove it. - Pick the paragraph style from a menu: headings 1-3, bullets, numbers, checklists, quotes. The one in force is ticked. - Insert a code block, a link, or a table — including the alignment row nobody remembers. - "More formats" lists all twelve on one screen, each with its name and a tick, so nothing has to be guessed from an icon. On iPhone there is an entry above the keyboard too. - All of it runs on your device: no tokens, no subscription, works in airplane mode. - For a layout you can't name, write a sentence instead — "turn these three paragraphs into a table" — and the AI comes back with a diff to approve. Quit and come back — your files are still open - Reopening the app puts back the tabs you had open, right down to the one you were looking at. Each workspace remembers its own. - Files deleted or renamed while you were away are skipped, instead of greeting you with a stack of errors. Tasks: run the whole thing - A task can now name a script. Applying it offers "Apply and run the script" — settings switched, opening message filled, the routine done. - If the script isn't in the current workspace it says so, rather than the button quietly not being there. Fixes - Long replies render far faster. Output used to slow down as it grew — writing the same content to a file was dramatically quicker than showing it in the chat. They now cost the same. - A change waiting for your approval scrolls itself into view instead of sitting below the fold, where the turn looked stalled. - Rotating the device during an AI illustration no longer loses it. That image had already been generated and already been billed — now it waits for you. - Files downloaded by a tutorial link open in the editor, and a script that arrives with one shows up in the ▶ menu immediately (it never runs on its own). - The Mac app icon now matches the iPhone and iPad one.

A KB library the AI can search itself, and everything you tune is now shareable KB library — file it once, let the AI look it up - File a PDF, Word, PowerPoint, Excel or Markdown document into the KB library and reference it as often as you like, instead of pasting it in again every conversation. - Type `~` in chat to attach one — or do nothing and let the AI search the library itself. It reads only the passage it needs rather than pulling the whole document into the conversation. - Ask in plain language. Search was replaced this release: it used to need a literal substring of the entry, and CJK questions almost never matched. Now "when are invoices due" finds the entry. - When nothing matches it says so, instead of handing back five things that merely look related. - Building a deck or a document can draw on KB entries too. Sharing — hand over what you tuned, or install someone else's - Prompts, export-style templates and KB entries export as files (.aiide-prompt / .aiide-style / .aiide-kb). - Browse and install from the built-in official library, or add your own source — any public GitHub repo or web page. No registration, no review, no gatekeeper. - Sources flag newer versions and update in one tap, and **a copy you edited is never overwritten** — you choose to keep it or replace it. - Language variants of the same item collapse into one row, with a switcher on the detail page. - Double-click a .aiide- file in Files or Finder to open it, and blogs or tutorials can carry one-click install links. - You always see the full contents before installing — especially for a prompt, whose text joins the front of every later turn of your conversation.

A new way to illustrate slides, saveable export styles, and a round of Mac polish Slides can be illustrated a second way - New "Image + text boxes": the AI paints only the background, and the title and bullets sit on top as native text boxes. Fix a typo in place instead of paying to regenerate the slide — and CJK never comes out mangled. - The original full-page image mode is unchanged. It can lay out things a text box can't, so both are in the menu. Export styles can be saved as templates - AI refine, auto illustrations and full-page images each keep their own list of named templates, one tap apart. - Get a style right, save it, and start the next deck without retyping it. Long-press (right-click on Mac) to rename, update with the current text, or delete. Mac polish - Fixed a second window opening when the app was launched from Xcode or from a link. - Fixed some interface text not following a change of language. - Finishing an export now shows a proper panel: filename, where it went, and Open / Show in Finder / Share. - Prompts can be deleted from the library — on Mac there was previously no way to do it at all. - Fixed the layout of the export, AI illustration, find-and-replace, workspace-switch and version-control panels; "file search" in particular used to look like it had frozen. Also improved - Fixed typing stalling while the @ file list was open in chat. - Fixed a rare case where an active subscription was asked to subscribe again. - Fixed some newer OpenAI models refusing to send once data tools were switched on. - Removed the media-gen skill, which no longer works — its API endpoints were retired upstream. For image generation use Settings → AI Illustration.

Mac app, Office and PDF editing, and an AI that can look things up AI-IDE on the Mac - A native macOS app: the same editor, preview, chat and version control, plus a full menu bar and keyboard shortcuts. - One subscription covers iPhone, iPad and Mac — you don't buy it twice. - PDF export works on the Mac too. Edit Office files and PDFs directly - PDF: open one to read it and send selected passages to the AI; reorder, rotate and delete pages, or save a selection as a new file. - Excel: edit cells in place. Existing formatting, formulas and other sheets are preserved. - Word: edit paragraph text without changing its style. - PowerPoint: edit the text on each slide; layout and fonts stay as they were. - Exported Office files are now compressed, so they are noticeably smaller. The AI can look things up itself - 30 built-in data tools: full-page web extraction, YouTube, academic papers, US financials and SEC filings, crypto prices, posts and reply threads on X, prediction markets, and Perplexity answers. - Switched on by category, with the per-message token cost shown next to it — turn off what you don't need. - The chat header switches tools on and off for the turn you're about to send. - Three ready-made research procedures are included: multi-source research, US equity analysis, and multi-signal trend forecasting. - The recommended-skills list is now limited to what the built-in tools can't do. Search, financial and social data that used to need a skill come straight from the tools instead — more reliable routes, and no instructions resent every turn. Tutorial links arrive ready to run - A tutorial link can specify which data tools to switch on. Before anything changes you see which tools each category brings, the estimated token cost, and anything that will be switched off. Also improved - Data limits are yours to set: API response size, file reads, attachments and the largest file you can open — with one tap to restore the defaults. - Fixed chat and editor losing their scroll position when the device is rotated. - Fixed some models returning HTTP 400 after a tool call. - Fixed applying an AI edit sometimes overwriting other parts of the file. - With several edits in one turn, the jump to preview now happens once at the end instead of mid-turn. - Added a tutorial link under Settings → About.

Subscriptions, prompt data requirements, and @file autocomplete AI-IDE Pro is here - The AI features (chat, AI-refine on export, AI illustrations) are now a subscription with a 7-day free trial. Editing, preview, export, and version control stay completely free. - Anyone who downloaded AI-IDE on or before 19 July 2026 keeps every AI feature free forever — detected automatically, no code to enter. Prompts can say what data they need - Declare requires: sales.csv at the top of a prompt. Send without referencing that file and AI-IDE warns you first, instead of letting the model invent plausible-looking, wrong numbers. If the file is in your workspace, one tap attaches it. - A requirement can be a role name (requires: sales data) that you bind in the message — "sales data@sell.md" — so a shared prompt needn't assume your filenames. One name can bind several files. See what a turn actually cost - The chat header shows the last request's input tokens as "tools / total", so you can tell how much of the bill is tool traffic. Slide exports no longer lose your work - Pick the folder and filename before exporting. The file lands in your workspace — in the file tree, version-controlled, reopenable — and can still be shared. - Full-page image decks resume: after an interruption the next export offers "continue from image 4", so finished slides are never paid for twice. - Fixed exports being cut short when the screen turned off or the phone was rotated. @file references without typing the whole path - Type @ in the chat for a file menu: open tabs first, then the rest of your workspace, filtered as you type. - The menu now appears wherever the cursor is, not only at the end of the message. - Filenames with CJK characters or spaces can finally be referenced. Interface and preview - Fixed dozens of places where the English and Simplified Chinese interfaces still showed Traditional Chinese — the preview's Annotate and Copy buttons, the outline, version history, retry banners. - Pinch to zoom in the Markdown preview. Image-only decks are readable by the AI - Attach a full-page-image PPTX or scanned PDF with almost no text, and AI-IDE offers to read its pages with a vision model. - Fixed images never reaching Gemini models. More reliable skills - Fixed every AISA skill failing to load with HTTP 404 after upstream moved them into category folders. Both layouts now work. - The recommended-skills catalog now lists only skills that can actually run here, and script-based skills are flagged on every import path. - Deep links accept an aisa:skill-name shortcut; the same skill no longer imports twice from different URL spellings. Demo scripts - New input_text (ask the presenter for a value mid-run) and screenshot (captioned feature shots) commands.

End-to-end media generation, custom export styles AI finishes long-running generation jobs (e.g. AISA media-gen) - New download_file tool: once a video/image/report finishes rendering, the AI saves the result straight into your workspace — after your one-tap approval, no more copying expiring download links by hand. - Safe by design: only links returned by your own APIs can be downloaded — the AI can never invent a URL — and every download shows a confirmation card with the source host and destination path first. - Smarter polling: the AI can wait a few seconds before checking job progress instead of burning turns on "still running"; the per-turn tool-call limit rises from 6 to 16, and hitting it now says so clearly (type "continue" to resume) instead of stopping silently. Custom export styles - New Settings → Export Style: describe in your own words how AI refine, auto illustrations, and full-page image slides should look — bullets per slide, deck length, tone, and art direction are yours to set. - 17 built-in examples to apply with one tap: dense / minimal / teaching / executive / story arc, watercolour / isometric 3D / line art / paper cut-out, dark tech / editorial / Japandi… fully editable after filling. - A Style shortcut appears in the export panel once AI refine is selected; leave fields empty and everything behaves exactly as before. Polish - Referencing a large file with @path no longer fails the conversation: only the first portion is attached, with a pointer telling the AI to page through the rest. - The prompt library and chat menu now sort alphabetically, so the list stays stable and scannable.

Token usage stats, one-tap import, portable scripts Token usage statistics - New Settings → Token Usage: cumulative input/output tokens broken down by provider and model, with totals and thousands separators. - Kept separately from chat history — deleting chats doesn't change the numbers; one tap resets everything to zero. One-tap import (deep links) - An aiide:// link from a tutorial or website can seed data files, prompt instructions, and a pre-filled chat all at once. - A confirmation sheet always lists the files, prompts, and their source hosts before anything runs; the chat is only pre-filled, never auto-sent. - After importing, a fresh chat opens with only the imported prompts active, so the run starts from a clean, predictable state. Portable scripts & smoother runs - New prompt command: carry prompt content inside the script (or your synced workspace) so it runs the same on another device, without depending on that device's Prompt Library. - Preflight: before a run starts, missing providers are reported clearly upfront instead of failing midway. - When a step fails, a Retry / Skip / Stop dialog appears — no need to re-run the whole script just because the last step hit a hiccup. Polish - After you approve an AI edit, the file opens in preview so you see the applied change right away. - When no AI model is set up, the chat proactively points you to Settings instead of failing only after you send. - Added a once-a-day "what's new" tip with a tutorial link.

Demo scripts, plus steadier editing and chat New: Demo scripts (.aiide) - Automate the app with a line-by-line script and reproduce a whole flow — create a file, chat with AI, annotate the preview, batch-edit, export a deck — as if an agent were driving AI-IDE on screen. - Choreography: automatic tab switching, typewriter input, a top step banner (pause / stop), and the screen stays awake for the whole run. - Cached offline mode: replace AI calls with pre-recorded content — zero network, zero tokens, identical every time — ideal for recordings and stage demos. - First run asks you to accept a risk notice; the editor syntax-highlights .aiide (commands / comments / strings / variables). Fixes & stability - Version control: fixed "phantom changes" — files reclaimed by iCloud (to save space) or deleted no longer keep reappearing as uncommitted changes. - Editor: fixed files starting with a "#" that couldn't scroll right after opening; scrolling and editing are smooth throughout. - Chat: quota / rate-limit / auth errors now fail fast instead of retrying pointlessly (which slowed things down and worsened rate limits). - Tool calls: fixed an "Unknown tool" caused by some providers' streaming, so AI tools are more reliable. - Continued polish for 0.9's file management, AISA Skills, and multi-prompt library.

Slide export, AI illustrations, AISA skills, and a multi-prompt library Export to PowerPoint (PPTX) - Turn a Markdown outline into slides in one tap; "AI Refine" distills a whole document into a deck, with cover / section / two-column / quote layouts. - Import your own .pptx as a template — exports reuse its slide master, layouts, and theme, so company decks come out on-brand. AI illustrations (bring your own image model) - Generate an illustration from one prompt (OpenAI gpt-image and Gemini image models), inserted at the cursor — it renders in the preview and flows into PDF / Word / PowerPoint exports. - For decks, pick "auto illustrations" (AI Refine adds images to suitable slides in a two-column layout) or "full-page images" (each slide rendered as one designed picture). Brand Kits & professional layouts - Build a Brand Kit (primary colour, fonts, logo, cover background): applied to PDF covers, Word headers, and every slide, for a consistent look across formats. - Four new layout modes for PDF / Word: Reading (larger type, loose leading, narrow measure), Report, Academic (section numbering + first-line indents), and Manual (accent-bar sections). - A unified export panel covers all four formats and custom templates, with live progress while the AI generates. AISA skills & unified gateway - New AISA provider preset: one key for GPT / Claude / Gemini / Grok / DeepSeek / Qwen and 70+ models. - Built-in AISA skill catalog: browse 41 official skills (web & scholar search, US stocks, crypto, prediction markets, X/Twitter, YouTube, SEO, media generation) by category with localized summaries; read the full SKILL.md before importing. - Chat gains a locked-down http_request tool — HTTPS-only, host-whitelisted, your key injected on device and never shown to the model — so imported skills can fetch live data and synthesize answers. - Enter the key once: set AISA up as an LLM provider and Skills plus web search reuse it automatically. Prompt library upgrades - Activate several prompts/skills at once — merged in your chosen order, with a combined-size meter and an oversize warning. - New search bar and origin groups (custom / AISA skills / imported); the chat-header switcher groups the same way. More flexible web search - Four AISA search backends added (web / scholar / hybrid / Tavily relay), reusing your AISA key. - Each search service now stores its own API key, so switching backends no longer overwrites keys. File management - The file tree's long-press menu gains Move to… (with a folder browser), Duplicate, and Share. Fixes & polish - PDF now paginates properly (no more one giant page); wide tables wrap instead of getting clipped. - Ordered lists in the preview no longer all show "1."; editor line numbers scroll with the text. - Fixed an intermittent editor scroll lock. - Missing or failed AI illustrations now report a clear reason before and after export instead of silently producing an image-less deck.

Slide export, AI illustrations, Brand Kits, and four layout modes Export to PowerPoint (PPTX) - Turn a Markdown outline into slides in one tap; "AI Refine" distills a whole document into a deck, with cover / section / two-column / quote layouts. - Import your own .pptx as a template — exports reuse its slide master, layouts, and theme, so company decks come out on-brand. AI illustrations (bring your own image model) - Generate an illustration from one prompt (OpenAI gpt-image and Gemini image models), inserted at the cursor — it renders in the preview and flows into PDF / Word / PowerPoint exports. - For decks, pick "auto illustrations" (AI Refine adds images to suitable slides in a two-column layout) or "full-page images" (each slide rendered as one designed picture). Brand Kits & professional layouts - Build a Brand Kit (primary colour, fonts, logo, cover background): applied to PDF covers, Word headers, and every slide, for a consistent look across formats. - Four new layout modes for PDF / Word: Reading (larger type, loose leading, narrow measure), Report, Academic (section numbering + first-line indents), and Manual (accent-bar sections). - A unified export panel covers all four formats and custom templates, with live progress while the AI generates. Fixes & polish - PDF now paginates properly (no more one giant page); wide tables wrap instead of getting clipped. - Ordered lists in the preview no longer all show "1."; editor line numbers scroll with the text.

Export to Office, fuller Markdown, and chat that survives flaky networks Export to PDF / Word / Excel - Turn a Markdown document into a PDF or Word (DOCX) file, or a table/CSV into Excel (XLSX). Generated on device, then shared via the system share sheet. Read more file types - PowerPoint (PPTX) and Excel (XLSX) attachments are now transcribed for the AI (PDF and Word were already supported). Markdown preview, closer to GitHub - Added [!NOTE]/[!TIP]/[!IMPORTANT]/[!WARNING]/[!CAUTION] alerts, ~~strikethrough~~, __bold__ and bold-italic, and bare-URL autolinking. - Headings are anchors: [text](#section) links jump within the document; backslash escapes and \| inside tables are handled. Resilient chat over flaky networks - If a reply stream drops, it auto-retries with backoff and a connectivity check, then continues — no need to tap Resume every time. - Repeated retries no longer balloon the conversation, so continuing uses less data and succeeds more often.

A fresh new look, plus a big upgrade to reading Markdown Two new themes Markdown preview, greatly improved - Document outline: tap a heading to jump, with your current section highlighted; shows section count and estimated reading time. - Smoother reading: a top progress bar, automatic "resume last position", and in-document full-text search (live highlight, next/previous). - Code: syntax highlighting, a language label, and one-tap copy. - Easier to read: wide tables can switch to cards, deep nested lists collapse, task lists show checkboxes, the file tree gains language badges, and VoiceOver support is improved. Lighter version control - Now versions only the files you open or edit — folders with thousands of files no longer get fully scanned or stall on "create version". Reset anytime from the version-control menu. Clearer and safer - The change (diff) card now shows line numbers and collapses unchanged sections, so you see exactly which line changed. - Important fix: edits could be saved into the wrong file after switching tabs — fixed.

New: mark up what to change right on the preview, and let the AI fix it all at once - Annotate directly in the Markdown preview: tap a block to mark a spot (e.g. "add an outline here"), or select text to mark a passage (e.g. "simplify this"). - Just type one instruction per mark; each mark is numbered on the document so you can see exactly what it points to. - Queue several marks (up to 5, adjustable in Settings) and send them together — the AI produces a single Diff you review and approve as usual. - Same safety model: every change is shown as a Diff first and only written on approval. Also: Traditional Chinese / Simplified Chinese / English manual with a language switcher; many localization and stability fixes.

Two big additions: local version control and a Prompt Library Local version control - Enable version control for any workspace, snapshot its current state as a version, view diffs, and revert a single file or an entire version. - History stays on device: never uploaded to iCloud, needs no git or GitHub account, and sidesteps the ".git in the cloud gets corrupted" problem. - The Files tab shows an uncommitted-changes count; optionally auto-create a version after each approved AI change. Built-in dedup + compression; clean up unused data with one tap. Prompt Library (AI instructions) - Build your own system-prompt presets and set one as the active global prompt to improve output quality, reduce hallucination, and lock in a style — switch it on the fly from the chat header. - Import community skills from a GitHub / raw link, or download from the built-in "Recommended Skills" catalog (coding, specs, business decks, sales proposals, internal comms, and more). - Privacy: the app ships only links; content is fetched from GitHub by your device when you tap Download. Also: Markdown preview now renders tables; new "Show hidden files" option; various stability fixes.

- Fixed Chinese input errors - Added Markdown file format preview to edit mode