Changelog

The main window is realigned, with one set of icon and button sizes.

• The main window uses the macOS 26 window style of the built-in apps: the same corner radius, and the traffic lights on the same line as the page title.
• The sidebar, page titles, and page content line up at the same positions, so nothing moves sideways when you switch pages.
• Icons, buttons, search fields, spacing, and corner radii use one set of sizes.
• In a narrow window, History details show the full time and status, and the Copy button shows only its icon.

The microphone opens when you press the shortcut, and the main window and history are faster.

• A single-modifier shortcut such as ⌘ alone opens the microphone on key down, so words spoken right after the release are kept. When the press is part of another shortcut, that audio is dropped. Bluetooth microphones still open after the release.
• History is stored in a SQLite database with faster search. The existing history is imported on first launch.
• The capsule closes as soon as the text is pasted. Restoring the clipboard and saving history happen in the background.
• The main window has a fixed-width sidebar and switches pages faster, and Tab only moves within the visible page.

Interface text is reworked in Chinese and English, and the English interface no longer shows Chinese messages.

• Interface text is shorter and uses one name per thing. For example, the Dictionary is split into Learned and Added Manually, and the balance messages say what to do next.
• With English selected, capsule errors, AI service and clipboard errors, and window titles all show in English.
• English counts are fixed, for example "1 dictation" and "Used 1 time".
• The level bars next to the microphone menu in Dictation settings are gone.

Home gets a flat background, and shortcut keycaps follow dark mode.

• Home drops the glow and the dot grid at the top; the background is now flat.
• Fixes shortcut keycaps at the top of Home staying white after macOS switched to dark mode while Lerro was running.

Onboarding and Settings are rebuilt, and AI cleanup takes three steps to turn on.

• Onboarding goes from eight steps to four: grant permissions, dictate once, choose whether to use AI cleanup, and learn the shortcuts.
• A three-step guide turns on AI cleanup: sign in to the DeepSeek platform, top up, then create a key and paste it, which connects automatically. It appears in onboarding, on Home and in Settings.
• AI cleanup supports DeepSeek only. The balance is read from DeepSeek and shown in the sidebar and Settings; when it runs out, Lerro writes the raw text and points you to top up.
• Settings go from six tabs to five: General, Dictation, AI Cleanup, Privacy and About. Usage details live in the AI Cleanup tab.

The dictionary learns on its own: names and terms that keep getting misheard are remembered.

• Names, products, companies, acronyms and jargon that the AI corrects in two separate dictations are added to the dictionary automatically. Each shows what it was misheard as and which dictations it came from, and can be removed.
• Home shows what Lerro learned this week; in History, remembered corrections are marked.
• Dictionary words record how often they are used, and the most used ones are sent to the AI.
• Find Words in History and dictionary snippets are removed.

The start sound no longer goes missing.

• With lowering other audio enabled, the start sound was sometimes silent after the Mac had been idle. Lerro now waits for the start sound to finish before lowering other audio; the microphone still opens right away.
• The “Stop automatically when you finish speaking” setting is removed. It relied on Apple's speech detector, which reports nothing in practice, so it never took effect.

New versions are easier to notice.

• Lerro checks for a new version every 4 hours. When you are using Lerro, the update window opens; when it is in the background, the menu bar icon gets a dot and the menu bar panel and main window sidebar show an update button.
• Downloading and installing still start only when you click.

See what the AI did, what it was sent, and what it cost for every dictation.

• History adds an AI section: the words the AI corrected and the fillers it removed, what it was sent (your words, app, dictionary, tone, and which context was not sent), and what the call cost with the calculation. The raw request and reply are one click away.
• Settings › Usage lists every AI call; select one to open its history entry.
• Settings › Intelligence shows the dictation, translation and dictionary-learning prompts section by section, with the examples as cards.
• A collapsed sidebar comes back with the button at the top left of the window or ⌃⌘S.

A new Home dashboard, and short sounds when a capture starts and stops.

• Home is a usage dashboard: total characters, time saved, speaking speed, dictation time and count, a year of activity as a heatmap with streaks, most-used apps and hours.
• A short sound plays when a capture starts and stops. Turn it off in Settings › General › Sounds.
• With lowering other audio enabled, other audio returns as soon as the microphone closes.
• Sidebar, Settings and menu bar icons use SF Symbols.

The website download is now a disk image: open it and drag Lerro into Applications.

• The website download is a signed and notarized DMG. It opens with Lerro next to the Applications folder on a branded background; drag it across to install.
• In-app updates work as before.

A lighter Lerro: the local AI mode is removed and the download drops from 11.7 MB to about 4 MB.

• The local AI mode (the Qwen model running on your Mac) and its 3 GB download are removed. AI polish and translation use the API you configure, such as DeepSeek.
• If you had chosen local AI, Lerro switches to Apple-only recognition. Your other settings stay the same.
• The download drops from 11.7 MB to about 4 MB.

A new brand and interface, better mixed Chinese–English recognition, and AI polish that fixes misheard words like an input method.

• A new brand mark: a cursor standing on a baseline forms the L of Lerro, with ink blue as the accent color. The app icon, main window, sidebar, and onboarding follow the new look.
• The dictation HUD is now a single ink-black glass capsule. Recording, live text, polishing, and failure notices all change inside the same capsule, and it fades out as soon as the text is written.
• The menu bar icon opens a new panel with the current status and recording timer, Dictate and Translate buttons, your last result, today's dictations, words and AI cost, and pickers to switch the microphone and intelligence mode.
• Speech recognition now uses Apple SpeechTranscriber. On 40 real recordings it recognized 62% of the English words in mixed Chinese–English speech, up from 39%, and no longer drops whole phrases.
• AI polish fixes misheard words the way an input method does: it replaces words that do not fit with words that sound the same or similar, and removes fillers. On recent dictations it fixed 6 of 8 misrecognitions, up from 2 of 8.
• The dictionary has a new Suggestions view. Lerro compares the raw and polished text of past dictations, lists likely recognition errors with how often they occurred, and you choose to add or ignore each one. Terms Lerro learns now apply in every app.
• The microphone opens before the speech analyzer is ready, so the first words of a dictation are no longer cut off.
• Settings has a new Usage page with remote AI tokens, requests, and estimated cost for today, 7 days, and 30 days, broken down by feature and model.

A focused voice-typing path with live preview, safe delivery, and vocabulary that learns.

• Press your shortcut to speak, follow a centered transcript that expands smoothly, then press again to write; the HUD disappears immediately after success.
• Strict app, input field, value, selection, and secure-input checks protect the captured target; failed deliveries remain on the clipboard with a compact recovery card.
• AI classifies useful corrections to recently dictated names and terms, adds them to the app-aware dictionary, and feeds the same vocabulary to Apple Speech and AI prompts.
• Personalization is now a first-level page with an app grid, real icons, per-app tone, enablement, and AI preview.
• The eight-step onboarding verifies real actions for privacy, Apple dictation, AI choice, shortcuts, first dictation, recovery, dictionary learning, and app tone.
• Apple dictation is the default, followed by remote AI and local AI; the 3.03 GB local download supports background progress, pause, resume, stop, and restart recovery.
• Ask, Rewrite, spoken follow-up edits, success receipts, and Apple Translation have left the active product; Translate uses the selected remote or local AI, while Quick Dictate remains optional and off by default.

A complete first run, from choosing AI to the first finished workflow.

• A nine-step guide teaches AI setup, permissions, shortcuts, Quick Dictate, receipts, spoken editing, and the full toolkit.
• Lerro recommends local AI or an API model from this Mac's chip, Metal support, memory, and free storage.
• OpenAI-compatible Provider, model, API key, context choices, and connection testing are available inside onboarding.
• Local model downloads continue in the background and support pause, resume, restart recovery, and explicit stop.
• Quick Dictate stays available through Apple Speech while local AI is still preparing.

Keep speaking to revise what Lerro just wrote.

• Quick Dictate starts with one Fn press and finishes after speech followed by a short silence.
• Say a follow-up to undo, delete a sentence, replace an exact phrase, or request a semantic rewrite.
• Every accepted edit stays bound to the same app, focused field, and delivered text fingerprint.
• A version chain supports repeated edits and restoring the prior version.
• History keeps the original delivery and its revision lineage for review.

Live, reversible voice writing with a controlled hands-free finish.

• The HUD shows the target app and transcript while Apple Speech is still listening.
• A focus-safe receipt offers Undo and immediate correction after delivery.
• Corrections preserve raw, processed, and corrected lineage and continue app-scoped learning.
• Saying “send it” or “发送” can submit in explicitly approved apps after a first-use confirmation.
• History records processing route, context sharing, and phase timings without storing new sensitive context text.

Context-aware voice writing that adapts to each Mac app.

• App styles keep mail, chat, documents, and code in the right voice.
• Command transforms any selected text with Fn Space or answers from the current context.
• Spoken snippets expand locally, while corrections teach app-scoped terms.
• A public 60-case benchmark keeps contextual accuracy and latency measurable.

Lerro now speaks English and Simplified Chinese across the app and its public home.

• The app interface now supports English and Simplified Chinese.
• The website and public documentation now offer matching English and Simplified Chinese routes.
• Release screenshots now ship in language-specific sets for both product surfaces.

Reliable Fn shortcuts that keep macOS Emoji out of the way.

• Built-in Fn and external Globe keys now use one physical-key ownership model.
• Every configured Fn sequence stays captured through its final release, including repeated modifier events.
• Shortcut monitoring now observes keyboard events only and keeps the same two-permission setup.

Faster shortcuts, deeper Apple Speech, and device-side translation.

• Fn shortcuts now use a cleaner two-permission setup and avoid the Globe menu.
• Apple Speech handles live revisions and audio endings more reliably.
• Translate now uses Apple Translation on the device, with clearer language-resource setup.
• Onboarding, shortcut settings, the idle HUD, and update downloads are easier to use.
• The website and GitHub now show the current product, real screenshots, and mirrored release artifacts.

The first public stable update.

• Automatic updates deliver signed, notarized releases inside Lerro.
• The public download is verified for macOS 26 on Apple silicon.