Performance and reliability polish for the core loop. • A snappier first correction — WunderType now prepares its history and personalization data the moment it launches, so your first correction of the session no longer pauses to load it. Saving to history also stays out of the way, even when your history is large. • Accurate On-Device status — after macOS reclaims memory from an idle on-device model, the menu no longer shows "Disconnected." Your next correction simply reloads the model and runs. • Smoother provider switching — switching away from On-Device AI right after launch no longer keeps loading a model you're no longer using.

New ways to connect WunderType — and new ways to trigger it. • Connect any OpenAI-compatible server — a new Custom Endpoint option lets you point WunderType at LM Studio, Groq, vLLM, Together AI, or a server of your own. Pick a preset or type the address yourself. An API key is optional, because most local servers don't need one. • Pick your model from a list — WunderType reads the available models from your server, so you can choose one from a menu instead of typing its exact name. • Right-click to correct — "Correct with WunderType" now appears in the Services menu in any app, so you can fix text without remembering a keyboard shortcut. • Shortcuts and Siri — two new actions, Correct Text and Transform Text with Prompt, let you build WunderType into any Shortcuts workflow or automation.

Look back at every correction — privately, on your Mac. • Correction History — a new History tab in Settings keeps your last 50 corrections. Expand any entry to see exactly what changed: your original text and the fixed version side by side, with the differences highlighted. • Copy anything back — grab the original or the corrected text from any history entry with one click, or select just the part you need. • Private by design — History is off until you turn it on, everything is stored only on your Mac, and one click clears it all. Nothing is ever sent anywhere.

A calmer progress indicator and a refreshed Settings window. • A quieter progress pill — while a correction runs, a small indicator now appears right next to your cursor, and only when there's actually a wait. Quick corrections stay completely out of your way, and a checkmark confirms when the text is ready • Clear feedback when a model is loading — if an on-device model needs a moment to warm up, the indicator now tells you instead of just spinning, so a slow first correction never looks stuck • Refreshed Settings — every section now has a clear, labeled heading, making the General, Prompts, Personalization, and About tabs easier to scan at a glance

Fresh on-device models and an easier way to manage them. • Two new on-device models — Qwen3 4B joins the lineup as a multilingual specialist (200+ languages), and the new Granite 4.0 Micro replaces Granite 3.3 with a small, fast model that now preserves your Markdown formatting • Manage your downloads — a new list in Settings shows how much space each on-device model takes and lets you delete the ones you don't need, so retired models never linger on your disk • Tidier prompt editor — when a custom instruction runs long, only the text box scrolls now, keeping the Save and Shortcut controls in place

Quicker mode switching and a more flexible local setup. • A shortcut for every mode — give Correct Grammar, Improve Writing, Make Concise, Make Formal, and Make Casual each their own keyboard shortcut, and jump straight to any of them from any app without changing your default mode first. They're in the menu bar too, under "Correct As" • Custom Ollama server address — point WunderType at Ollama on a different port, through a proxy, or on another machine, instead of only the default local address • Ollama connection status — Settings now shows at a glance whether your Ollama server is reachable, so a wrong address is easy to spot

WunderType now learns how you like your text. When you keep changing a correction back, it remembers and stops making that change — and everything it learns stays on your Mac. • Learns what you change back — when you repeatedly undo the same correction, WunderType notices and stops making it. Switch it on in Settings → Personalization, or when you first set the app up • Everything stays on your Mac — it only ever remembers short word rules, never the text you write, and nothing is ever uploaded • See and control everything it learns — the new Personalization tab lists every rule in plain language. Turn any one off, edit it, delete it, or clear everything with a single click • Add your own rules — set exact word swaps that are always applied to every correction, independent of the AI model — for example, always change "utilise" to "use" • Off until you turn it on — you decide whether to enable it; nothing is learned unless you opt in

A stability and performance release — lighter on battery and memory, quicker to respond, and more careful with your data. No new features; everything you already use just runs better. • Lighter on battery — cloud providers (OpenAI and OpenRouter) now check their connection far less often in the background, with an instant status refresh the moment you open the menu • Lower memory use — On-Device AI now releases its model from memory automatically when your Mac runs low, then reloads it on your next correction • Your clipboard stays private — while a correction runs, the temporary clipboard contents are marked private so clipboard-manager apps don't record your text • More reliable API keys — saving an API key is now safer, with no moment where it could be lost, and WunderType tells you if a save ever fails • Snappier menu — the menu bar menu and your custom prompts open faster • Cancellable downloads — you can now stop an On-Device model download that's in progress

Downloading an On-Device model now shows real progress — and WunderType tidies up after itself. • Accurate download progress — when you download an On-Device model, the progress bar now climbs with a live percentage as the model downloads, instead of a spinner that looked stuck • Automatic disk cleanup — interrupted or cancelled model downloads no longer leave large temporary files behind; WunderType reclaims that space automatically

Watch your corrections appear. A new experimental option types the corrected text into place as the On-Device model writes it, instead of replacing your selection all at once. • Streaming corrections (experimental) — switch it on in Settings → General → Experimental, and On-Device corrections are typed into place live, as the model generates them • On-Device only — streaming uses the local MLX model for instant output; OpenAI, OpenRouter, and Ollama keep the standard one-shot replacement • Works in more apps than before — because the text is typed in like a keyboard, the streaming effect reaches apps where live editing previously wasn't possible • Keep the target app in front while text streams in — switching away stops the correction, and ⌘Z undoes any partial text • Accessibility and polish — clearer VoiceOver descriptions throughout onboarding and provider setup, plus a lighter, smoother progress cursor while corrections run

On-Device AI is here. Correct text locally on your Mac — no Ollama, no API key, no internet after setup. • New “On-Device” AI provider runs models directly on your Mac’s GPU using Apple’s MLX framework • Download a model once, then correct text entirely offline — your text never leaves your Mac • Three models: Granite 2B (fast, ~1.3 GB), Phi-4 Mini (balanced, ~2.0 GB), or Gemma 3 4B (best quality, ~3.0 GB) • Zero setup — no external apps, no API keys. Select On-Device during onboarding and go • Requires Apple Silicon (M1 or later)

Onboarding improvements based on user feedback. - You can now choose your keyboard shortcut during onboarding — no need to dig into Settings if the default conflicts with your apps - Ollama detection is faster and clearer — if Ollama isn't running, the setup screen now shows a live indicator so you know it's still checking - The completion screen no longer cuts off the "Start Using WunderType" button on smaller windows - Fixed the onboarding progress bar jumping when using the OpenAI provider path

A small reliability release for setup and custom prompts. - Your "Install starter prompts" choice during setup is now respected — opting out no longer installs them anyway - Custom prompt shortcuts you add or change now work immediately, without restarting the app - Smart Selection now puts your cursor back where it was when a correction makes no changes, instead of leaving the whole paragraph selected (requires non-sandboxed build)

A quick fix release for custom prompt users. - Fixed a bug where updating from v1.3.x could cause existing custom prompts to disappear - Custom prompt keyboard shortcuts now pick up edits immediately — no restart needed - Custom prompts now show a "No changes made" HUD when the AI returns your text unchanged, instead of failing silently - Brief custom prompt instructions are now automatically framed for better AI compliance - Improved VoiceOver labels for prompt editing buttons

The biggest WunderType update yet — your writing, your rules. • Writing Context — teach WunderType your preferred names, terms, and style notes. They're applied to every correction, so "WunderType" never becomes "wundertype" again. • 8 ready-to-use prompts — Email Polish, Slack Reply, Translate to English, TL;DR, Bullet Points, Prose from Bullets, LinkedIn Post, and Commit Message. • Ollama keep-alive — a new toggle keeps your local model loaded in memory, so every correction is fast. • Redesigned Settings — clean tabbed layout with General, Prompts, and About tabs. • Slimmer menu bar — just your prompts and the correct action. No clutter.

A code quality and accessibility pass across the app. - Fixed a potential crash in the correction HUD when displaying error messages - Copy buttons in Ollama setup are now accessible to VoiceOver - HUD text now respects your Dynamic Type size preference - Ollama setup polling uses modern lifecycle management for better reliability - Consistent font styling across all onboarding screens

We're constantly working to make WunderType better for you. - OpenRouter is now available as a provider during onboarding — pick from hundreds of models on first launch - Updated brand colors across the app for a warmer, more polished look - Small bug fixes and improvements

A focused fix release — error popups now tell you what went wrong instead of showing a silent warning icon. - Error HUD now displays a clear message (e.g. "No text selected", "API key not configured", "Ollama is not running") instead of just a warning icon - Error HUD stays visible longer (2.5s) so you have time to read the message - Previously, error details were only shown via macOS notifications — if notification permissions weren't granted, you'd see a blank warning triangle with no explanation

A polish release that rounds out the v1.3 series with better onboarding, smarter corrections, and accessibility improvements. - Onboarding now asks for notification permission before accessibility — so you'll see error messages from day one - If the AI returns your text unchanged, WunderType skips the replacement entirely — no more cursor flicker on already-perfect text - The HUD and cursor spinner now respect Reduce Motion for users with vestibular sensitivity - Unusually long AI responses are now truncated automatically to prevent runaway pastes - Updated app icons and logo with refreshed color scheme

A quality release focused on smarter corrections. The AI follows your instructions more reliably, and the Settings screen is clearer about what each mode actually does. - Tightened correction prompts: markdown formatting, code blocks, brand names, and the language you wrote in are now reliably preserved across every mode - Make Concise reliably trims at least 25% — no more "shortened" results that came back the same length - Settings → Mode picker now shows what each mode does in plain language, instead of leaking the internal AI prompt - Improved reliability of the direct text-replacement path in native Mac apps

An accessibility and code-quality update. - Provider cards in onboarding are now proper buttons, so VoiceOver announces them correctly and reads the selected state - Onboarding screens now respect your text-size preference — icons and pickers scale with Dynamic Type - Modernized internal notification and concurrency code for better reliability under macOS 15 - Refactored the onboarding and HUD source files into smaller, single-responsibility modules

A quality and polish update that makes WunderType leaner and more reliable. - Refreshed Settings design with a cleaner, more spacious layout - Fixed an issue where correcting with no text selected could process unrelated clipboard content - Internal code improvements for better maintainability

We're constantly working to make WunderType better for you. This update brings OpenRouter support, giving you access to hundreds of AI models through a single API key. - New AI provider: OpenRouter — use Claude, Gemini, Llama, Mistral, DeepSeek, and hundreds more models - Choose any model by entering its ID in Settings - All API keys stored securely in macOS Keychain

We're constantly working to make WunderType better for you. This is a small but meaningful update that strengthens WunderType's security posture and makes it easier to stay connected. - Added quick links to changelog, website, and contact in Settings - Under-the-hood security improvements

We're constantly working to make WunderType better for you. This update focuses on reliability and security, making WunderType more robust for everyday use. - Fixed an issue where the correction shortcut could trigger twice - Improved security for API key storage - Crisp, high-resolution logo on Retina displays - General stability and reliability improvements