What’s new
in Lumora.
What changed in each version of Lumora, newest first. The app downloads new versions itself and installs them when you restart it.
- Lessons use a black workspace with larger text. You can turn on background effects in Appearance.
- Practice questions take up less space. You can open the optional confidence control when you want it.
- Subject maps open in learning order, and the end-of-lesson summary has been redesigned.
- Home, Library and Account explain your next step more clearly. Sound feedback has been updated.
- The first-run terms explain what goes to the AI provider and link to Privacy and Acceptable Use.
- The tutor is instructed to offer hints, feedback and similar examples when you ask for help with graded work.
Earlier in 0.4
- Many more pictures in lessons: every teaching step with something to show now gets one, in maths and in subjects like history and biology.
- Fractions, number lines, graphs, bar charts, angles, steps in order, cycles, timelines and dot grids are drawn instantly and exactly, right in the lesson, with no waiting and no extra credits.
- Other pictures are drawn by a faster drawing model: they arrive in seconds instead of up to a minute, and come out cleaner and better labelled.
- The same picture is no longer shown twice in a row.
- Pictures in your lessons can be enlarged with a click; Escape closes them.
- Text fields show a clear outline when you reach them with the keyboard, and the smallest labels are a little larger.
- The language setting is honest about what it changes: lessons, quizzes and the voice use your language, while the app's menus are in English for now (Croatian is partly translated).
- The Library is simpler. It lists your subjects, grouped in folders if you make any; starting a lesson is what adds a subject, so there are no create buttons to puzzle over. Each subject shows its progress and a Continue button for its unfinished lesson.
- One way to group: folders hold subjects. They are sections in the Library and groups in the sidebar, and your old sidebar groups became folders. A subject's ⋯ menu has Rename, Move to folder (where new folders are made), Merge with another subject and Delete.
- Duplicate subjects can be merged: the map, lessons and files of one move into the other.
- One word per thing: Lessons are your conversations with the tutor, the Summary is the written explanation of an idea, and Transcript shows what was said in a lesson. A subject has Map, Lessons and Files; an idea has Summary and Tutor.
- A subject's map draws itself the first time you open it. If that fails, it says so plainly and offers Try again. Ideas you have not reached yet say "Comes later" (with a clock, not a padlock), since you can still study them, and the main button is "Learn this with the tutor".
- New goals join the subject they belong to: "Fractions for tomorrow" goes into Fractions instead of becoming a new subject, and Home shows where a lesson will go before you start, with a way to change it.
- After a lesson, Lumora asks once when to nudge you about reviews. Pick a time or say not now; it only reminds you when something is due. Home shows the next review date, and a review that is due sits at the top of Home.
- Home and Library agree on your progress: a subject is in progress as soon as you start a lesson, with when you last studied it.
- Check yourself opens in the middle of the screen instead of a narrow side panel.
- Lessons name themselves after what you studied. Right-click a lesson in the sidebar to rename it or move its subject into a folder.
- A clearer lesson plan panel: where you are, how much you have learned and the next step. Plan and Notes also open beside the Summary tab.
- Notes look like notes: a lined pad, newest first, with notes from the same minute kept together. The pin sits in each reply's own row of actions.
- Ending a lesson asks once, then shows an animated summary of what you learned.
- The microphone wave follows your voice. Lumora waits longer before sending what you said and tells you when it cannot hear you. Shift+Space works while you type in the message box (a quick "I am" stays a space) and dictates into "Why did you pick this".
- Maths fits its column: long equations get a little smaller or wrap instead of scrolling sideways.
- Diagrams stay readable: labels and lines too dark for the page are recoloured automatically.
- A redesigned tutor picker, with a voice preview for each tutor.
- Optional quiet study music, and a level-up sound when a step is learned. The focus timer's sound can be turned off.
- Hiding the text while the tutor reads aloud now hides the question too, until the tutor has finished reading.
- Read-aloud says f(x) as "f of x" and no longer stumbles over maths.
- When a reply runs too long or the tutor fails, the message says so instead of asking you to check your connection.
- Sign in with one button: Continue with Google. The page the browser shows afterwards now looks like Lumora. The pause on the terms screen is shorter.
- Smaller fixes: the confidence slider reaches 100%, clicks no longer sound twice, the first-run Start button waits for a goal, and the white outlines around panels are gone.
- Fixes Lumora closing at launch with "Cannot find module 'fs-extra'" on a fresh install of beta.6. The installer had left out the parts of the update checker, so the app could not start (and so could not update itself). If you installed beta.6, please install this one over it.
- The installer is now checked for missing pieces before it is built, so this cannot happen again.
- The voice and speech engines no longer see sign-in keys or secrets from your computer, and only the Python that ships with Lumora is used in an installed copy.
- Only Lumora's own window can talk to the app's inner workings.
- The lesson screen holds still: your latest message or answer sits at the top, the reply appears under it, and nothing scrolls while it is written. "Back to the current question" returns to it.
- One question per step, with a picture above the question in the first teaching steps.
- The "+" menu in the lesson box attaches a picture (file or paste) as well as a PDF; pictures are for Pro and testing accounts.
- One read-aloud speed for the whole app: the speed button and the Voice panel slider change the same setting. Pausing read-aloud stays paused.
- A bug icon in the title bar reports what happened, with an optional screenshot that stays on your computer.
- The getting-started tour starts silent and plays 60% quieter.
- New glass sliders in the feedback form; calmer sign-in button and "Forgot password?" panel; Settings switches centre their selection.
- Tutor replies have a length cap and long lessons are summarised in larger steps for steadier, cheaper replies.
- Unused credits roll over for three months.
- The live voice circle is removed; read-aloud and speech input are unchanged.
- Fixes: renaming a subject no longer makes Resume open an older copy; a failed credit write is retried instead of being billed as the full hold.
- When your sign-in ends on the server (you changed the password elsewhere, signed out everywhere, or the account was removed), Lumora signs out and says why on the sign-in screen. Before, it showed "Error invoking remote method ... Your session has ended".
- App error messages no longer start with "Error invoking remote method".
- The Account panel stays readable over a lesson: it uses the same clear glass as Settings instead of showing the lesson text through it. Credits read in plain words first; model calls and tokens are under Usage details.
- "Hand off to another AI" is also in the Account panel, and the button on each subject page has an icon.
- Quick voice controls: the unlabelled checkbox is now a "Read tutor replies aloud" switch.
- "Open Lumora" on the website just brings the app forward when it is already signed in, instead of opening another browser tab.
- Account security: a late Google approval can no longer sign you back in after you signed out, and an older sign-in or reset request cannot adopt a newer password.
- Messages about credits no longer say you are out of credits when a request only needs more available credits.
- Lumora moves to lumoratutor.app. The app talks to api.lumoratutor.app and opens sign-in and terms on www.lumoratutor.app. Earlier installs keep working on the old address.
- Credits follow real AI cost. One credit is US$0.001 of what the AI provider charges for a request, and credits never expire. A request holds at most the credits you have left, so a nearly empty balance can still finish a reply instead of being refused.
- New accounts get 150 free credits once their email address is confirmed. Google sign-in confirms it straight away. Confirming email, resetting a forgotten password, asking for more credits and deleting the account all work in the app and on the website.
- Replies stream as the tutor writes them in cloud mode, so text and read-aloud start sooner.
- PDF study material follows the plan the server reports. Beta.3 locked it for everyone.
- A notice appears when credits run low, with a way to ask for more.
- After a few minutes of use, once per version, Lumora asks for feedback: how likely you are to recommend it, plus ratings for tutoring, read-aloud, listening, design, sound, accessibility, speed and features. It reaches the team's inbox. Settings > Send feedback works at any time.
- Crash reports are opt-in. Lumora asks before sending one, or follows your choice in Settings (Ask, Always or Never). Reports carry no lesson content, and personal file paths are removed.
- A six-step tour after setup points out the goal field, setup, Start, Library, Settings and Account.
- New exports. "Hand off to another AI" on each subject sums up what you have mastered, your gaps (with the exact mistakes) and what is untested, written for another AI tutor. "Save as study notes" turns a lesson into clean Markdown notes, without the quizzes and chat.
- Security fixes from the 29 October audit. Signing out or switching accounts stops the previous account's lesson work and background jobs. A crafted backup can no longer overwrite or expose other lessons' files. Reminders only use the active account's subjects.
- Requests for DeepSeek Pro are answered by DeepSeek V4.1 Flash.
- Cloud tutor calls no longer retry a failed or uncertain generation automatically. Every logical request carries a unique request key, preventing transport retries from becoming separate billable requests.
- Stop and request deadlines now reach the underlying HTTP operation. Already-started provider work can still consume credits; cancellation is not a promise of a refund.
- Insufficient-credit, pending-access and uncertain-usage errors give account-specific next steps instead of asking the learner to debug the connection or blindly retry.
- Browser sign-in can return to the desktop through a registered Lumora link. Short-lived, one-use approvals are bound to the app's PKCE request; account passwords and reusable sessions are never placed in links.
- Website navigation keeps Account simple and Download visible. The account page handles app-opening prompts, expired links and retries without claiming the app opened before it can know.
- Sign-in methods cannot submit simultaneously. Setup saves once, startup status remains usable under React StrictMode, and cloud account failures have retry/refresh controls.
- Account credits use their actual cloud units and beta access state. The profile dialog remains closable at 200% zoom.
- Unanswered quiz questions remain visible in a small window, and scrolling back through lesson history keeps your place.
- Speech recognition uses its own authenticated local engine port and recovers cleanly from startup or engine failures.
- New broad lessons include an answerable first question. A missing question gets one repair attempt, and failed openings can be retried without losing saved lessons.
- Closed two backend races affecting expired monthly allocations and session creation during password changes or suspension.
- Native runtime checks cover account creation, consent/setup, first lesson, browser handoff, encrypted credentials and remembered sign-in. Live service checks use disposable accounts. Human approval and a second-machine test remain required before a wider launch.
- Private-beta Windows build remains unsigned. Paid checkout remains disabled; beta access and credits are assigned by an administrator.
Older versions0.1.4 to 0.3.1
- Google sign-in now works in installed builds. It previously worked only on a machine with the OAuth client secret in its own
.env: the packaged app has no .env, so every fresh install answered "Google sign-in is not set up yet". The secret is now baked into the main process at build time from the build machine's .env, and never enters source control or the renderer.
- The app wears the Luna skin: near-black instrument surfaces, layered glass with curved reflective edges, a quiet aurora behind the workspace, and the same type pairing the website uses.
- Home is built around resuming. Continue restores your last saved lesson, one goal field starts the next one, setup stays folded until you want it, and due reviews plus your real subjects sit in the same column.
- Library opens with every subject and optional folders. Notes and materials live inside a subject, and search spans the whole library.
- Study returns to a reading column: centered lesson and composer, an optional tutor and path panel, a compact voice orb, focus mode, and a focus timer.
- Lumora Cloud accounts: sign in with email and password or Google, and your subjects and preferences belong to that account on this machine. Paid checkout is still closed.
- First run has room to breathe: a longer Luna splash, a reading pause before accepting the terms, and a searchable country picker with real flags.
- Under the glass: credentials move to OS encryption, voice synthesis spawns behind per-process authentication on ephemeral ports, PDF parsing runs in a worker, and prompt context is bounded by journaled snapshots that survive a quit.
- Skill tree is now a constellation: every concept is a star sized and lit by its mastery, branches gather as constellations, the up-next star pulses. Labels stay readable from 3 concepts to 100.
- Mastery moves in smaller steps (one correct answer no longer jumps 0% to 25%), and sessions and lessons finally share one formula.
- Steadier generation: prerequisite edges survive forward references in the same reply, and thinking traces no longer swallow the real tree.
- Lessons survive renames: continuing a subject restores its transcript even when the subject was renamed after the snapshot.
- Home start form ships three example starters; the subject box is a real dropdown; voice skip is one click; the tour no longer ambushes returning users.
- Voice-model gate on first run, session pins strip, push-to-talk VoicePill composer mic, splash wait-skip, tabbed settings with Licences tab.
- Hide-the-lesson button on every tutor message: blur the explanation and answer the quiz below from memory, then reveal.
- The update checker now spins a Windows-style waiting ring instead of a dead button, with a circular-arrows icon on the Check button.
- In-app updates: new versions download themselves (small differentials, not the full installer) and install on restart from Settings → App updates. No more redownloading by hand.
- "I don't know" is now a third quiz outcome: it never moves mastery and never returns as a mistake to correct (with a database upgrade for existing installs — old answers keep their meaning).
- Voice playback rebuilt on a re-synthesis scheduler: gapless blocks that survive stalls and burst traffic, plus a faster Low quality default.
- Skill tree is now a radial map that branches from the middle, with mastery rings on every concept; home cards no longer dilute mastery with branch nodes.
- The microphone engine now warms up on the loading screen, and speech fades in softly instead of popping.
- App accessibility pass: real labels, skip link, keyboard voice control.
- Slim Windows installer (257 MB) with the Python voice runtime inside; voice models download once on first launch.
- Gapless chunked voice playback plus engine warmup at boot.
- Standby mode (hide to tray) and opt-in testing-phase telemetry consent.
- Home refresh; tutorial language step with live voice preview and volume.
- The probe → plan → teach method credited to Eero Alvar in-app and on-site.
- First public Windows installer with bundled voice runtime.