VibeTutor changelog

Ordinary typed Live messages no longer repeat conversation policies, voice settings, and call-mode explanations. Exact timer checks remain available when asked.

First and returning calls use short opening briefs without duplicating tutor profiles. Passive classroom updates send compact change notices; full content is still read before editing or reviewing.

Fixed the actionEngine object-is-not-extensible error that blocked Flash-Lite and Auto classroom updates. Engine metadata now uses a separate request copy while keeping progress callbacks and queued preferences intact.

Each classroom tab now has Auto, Live only, and Flash-Lite only action engines. Preferences are remembered per account in this browser. Live can apply validated content directly; delegated work and retries stay on Flash-Lite, without Flash or Pro text-generation fallbacks.

New chat messages show sending, queued, and sent status with a quiet activity indicator and elapsed waiting time. Live delivery follows the actual socket queue, and indicators clear on replies, errors, or call end without adding read receipts or transcript noise.

Malformed Whiteboard plans, invalid SVG, and incomplete tool responses now enter bounded retries with visible retry and fallback progress.

Shorter, destination-specific AI instructions preserve the original learner request and remove conflicting read-versus-edit and Canvas runtime guidance.

Document and HTML edits can recover from stale revisions using a fresh read while preserving intervening learner changes.

Background workers now protect newer results from stale attempts, renew active leases, and require browser-confirmed application before completing a job.

Removed repeated internal instructions from classroom tool responses so tutors receive compact private status updates instead of paragraphs that could be spoken aloud.

Consolidated classroom conversation guidance into one short prompt while preserving tutor personalities, learner-selected tabs, and verified completion checks.

Canvas generation now receives only its selected engine instructions. Blocked browser globals and other code compatibility errors get one targeted regeneration with clear progress while existing drawings remain protected.

Typed and microphone calls now resume recognized provider aborts even shortly after a connection refresh, preserving the active lesson and queued typed messages instead of stopping solely because the replacement connection is new.

Added a clear portrait play action and a responsive, keyboard-accessible video dialog with standard playback controls.

Tutor video media is reusable through profile media records, includes graceful asset fallback, and adds VideoObject metadata for public profile discovery.

Added an experiment builder for choosing one tutor, one AI student, a lesson goal, and a five-to-thirty-minute session.

Tutor and student pickers now include searchable, keyboard-friendly profile results for quickly filtering large directories.

Two distinct Gemini Live sessions now take serialized spoken turns while a dedicated transcript shows who is speaking and what classroom work is underway.

The tutor can use the existing verified classroom tools while the AI student remains a bounded learner; pause, duration, repetition, turn-limit, and error guards keep the experiment controlled.

Observer work uses separate browser storage and does not write tutor memory, learning paths, AI-student mastery, or account homework.

Corrected the observer controller window binding so its isolated classroom runtime initializes reliably instead of stopping during page startup.

Tutor and AI-student transcripts now type live from the first spoken words instead of waiting for voice playback to finish, then settle into fully formatted Markdown.

Added durable background generation for Whiteboard sketches, images, and SVGs plus Canvas, HTML, Document, Code Editor, Quiz, and Homework updates.

Classroom activity now shows elapsed progress, provider retries, fallback use, browser application, and final verified readiness without blocking Gemini Live.

Unfinished work reconnects after a reload, remains ordered within its original lesson tab, and applies without switching the learner away from the tab they are using.

Added one authenticated browser-assisted worker per queued action and stopped status polling from repeatedly spawning PHP processes, preventing shared-host worker storms and storage contention.

Corrected SQLite immediate-transaction handling on PHP 8.1 and LiteSpeed so queued classroom jobs can be claimed without a false "There is no active transaction" failure.

Browser validation still protects saved work and can automatically regenerate rejected Canvas syntax, unchanged HTML, or malformed Document SVG content before reporting success.

Administrators can enable or disable background classroom generation from Classroom capabilities, while worker usage remains attributed to the correct learner and tutor.

A dedicated AI Students directory launches with four original learner profiles covering coding, workplace English, health science, and practical mathematics or physics.

Private teaching sessions support explanation, misconception diagnosis, application, and question-led practice with account-saved transcripts and topic-specific learner progress.

Each response separates personality from knowledge state and reports demonstrated understanding, remaining gaps, confidence, a concise teaching note, and a useful next probe.

Progress increases only when the simulated learner paraphrases, predicts, calculates, or applies the idea, rather than merely agreeing with the human teacher.

Selected places can now show a compact Google Places card with available photo attribution, address, category, rating, status, opening hours, accessibility, local time offset, phone, website, and Google Maps link.

A themed Layers menu adds Road, Satellite, Hybrid, and Terrain base maps plus independently switchable traffic, transit, cycling, and rich place-detail controls.

The Map can switch between 2D and lazily loaded Photorealistic 3D while preserving the lesson location and mirroring lesson pins between views.

Country, state or region, and city polygons can be switched independently in 2D and selected for place context when a compatible Google vector Map ID is configured.

Tutors can build compact sourced choropleth comparisons for two to twelve regions, with deterministic data-driven colors, formatted values, and a visible source legend learners can hide or restore.

Gemini Live receives the verified 2D/3D mode, camera, overlay preferences, selected place facts, and pin state and can operate those controls directly without a Flash-Lite delay.

Selected videos now load runtime, creator, publication, language, caption availability, quality, category, public engagement, tags, topics, and other available YouTube metadata.

Creator-supplied description timestamps become compact chapter controls, with exact current time, duration, player state, playback rate, progress, and current chapter available to the tutor.

Learners can mark the current moment for discussion, while tutors can prepare, seek, play, or pause verified videos only when the learner explicitly requests the playback action.

The official YouTube IFrame Player API powers live watch state and chapter navigation while privacy-enhanced playback and ordinary iframe fallback remain intact.

Added keyless Project Gutenberg search, complete sectioned reading, in-book phrase finding, text sizing, and explicit section sharing with the tutor.

Google Books and Gutenberg retain separate state, storage, tool calls, and controllers beneath one space-saving Books tab, so either provider can evolve independently.

Gutenberg catalog and text access is authenticated, rate-limited, HTTPS allowlisted, size- and time-bounded, cached locally, and delivered to the classroom one readable section at a time.

Tutor routing now follows the learner-selected book source and reads its verified current state before searching, opening, navigating, discussing, sharing, or clearing content.

Opening a Gutenberg result now prefers direct HTTPS text files and retries verified alternate editions when a catalog URL redirects to insecure HTTP.

The full multi-paragraph tutor biography and What you can explore focus list remain visible, alongside a compact learner-fit snapshot and a consolidated lesson-details section.

Personality and teaching-style traits are merged without duplicate labels, while lesson plans, classroom tools, learning paths, reviews, and public activity remain easy to find.

Course directory pages, course detail pages, and their course-image entries are no longer included in the XML sitemap while that content is being improved.

Place search, visible Map reads, road or satellite view changes, zoom, coordinates, and explicit pin clearing now run directly from Gemini Live for lower classroom latency.

Direct Map actions retain progress feedback, strict argument schemas, learner-intent safeguards, browser-render verification, active-call continuity, and concise tutor confirmation.

Flash-Lite remains responsible for generative and stateful classroom work, and the Admin System page now distinguishes direct Map functions from delegated classroom actions.

Focused YouTube searches, verified-result selection, current-state reads, and explicit clearing now run directly from Gemini Live without a Flash-Lite planning delay.

Complex YouTube curation, comparison, ranking, sequenced watch plans, and ambiguous requirements continue through Flash-Lite with a fresh verified result-state read.

Direct YouTube actions can request one semantic Flash-Lite fallback for query refinement or verified-result selection, while authentication, quota, network, and clear-intent failures stop without a wasteful model retry.

Learners can point to or tap any Map location and explicitly ask the tutor to pin it; the direct Live tool uses that exact browser coordinate and returns available area details for follow-up discussion.

The classroom header and workspace tabs now use a denser, scrollable layout so more tools remain visible while preserving more vertical space for the active lesson.

Screen sharing is now a separate Admin-controlled classroom capability and is disabled by default; when off, its control and preview are removed entirely.

Added an immersive Books classroom tab with verified Google Books search, compact reading suggestions, and the official Embedded Viewer for available previews.

Tutors can inspect the Books shelf, search by title, author, ISBN, or topic, open only verified volume IDs, and help learners navigate pages, zoom, and highlights.

Book previews respect edition, regional, and rights availability; unavailable previews retain verified metadata and a Google Books link without implying full-book access.

Google Books previews now mount at the full reader size, refit after classroom layout changes, and only open editions Google explicitly permits for embedding.

Live tutor greetings are now single-use per connection: classroom tool results remain active-call continuations, and an accidental post-action welcome is stopped and retried without restarting the lesson.

Added an immersive YouTube classroom tab with focused educational search, a large embedded player, compact lesson suggestions, and direct learner playback controls.

Tutors can read the current YouTube state, search for real embeddable videos, prepare a verified result without autoplay, and clear suggestions only when the learner asks.

YouTube uses strict SafeSearch, brief query caching, privacy-enhanced youtube-nocookie.com playback, visible action progress, and a distinct tutor voice cue.

Added a compact Map classroom tab with place search, lesson pins, and road, satellite, hybrid, and terrain views.

Tutors can inspect the visible map, locate a place, adjust the view, and clear pins through state-aware classroom tools.

Map actions now show visible progress, use a distinct tutor voice cue, and report success only after the requested view is rendered.

Added HTML beside Canvas with a responsive iframe presentation, optional syntax-highlighted source editing, starter content, undo and redo, download, and local persistence.

Gemini Flash-Lite can read, create, restyle, and clear complete HTML presentations through revision-bound classroom tools.

Generated HTML is staged before saving and runs in an opaque iframe sandbox without same-origin access, forms, popups, top navigation, device permissions, or access to VibeTutor storage.

HTML is now presentation-first: the rendered lesson uses the full workspace width while source remains collapsed in a small optional disclosure below it.

Tutors in calls and chats started without Classroom now explain how to restart with Classroom with Whiteboard selected when a learner requests a classroom tool, without interrupting ordinary conversation.

Classroom tab switches now stay local to the interface and no longer send automatic Gemini Live turns; automatic lesson-plan synchronization also uses the non-speaking context channel so a new lesson produces one greeting.

The classroom HTML tab now renders inline and display LaTeX with sandboxed KaTeX support, including formulas added by tutor-generated presentations, and its navigation label is shortened to HTML.

Live calls now show a Stop tutor control while speech is playing, allowing learners to take the floor without ending the call, timer, or classroom session; microphone barge-in also clears buffered tutor audio immediately.

The HTML classroom now gives nearly all available height to the rendered presentation: duplicate headings and the status footer are merged into one slim rail, source editing opens as an overlay, and Focus mode reduces controls to one restore button.

Community tutor portraits now use the same Gemini image pipeline as Classroom, with Nano Banana 2 and Nano Banana 2 Lite available in the creation form.

Generated portraits are validated and saved under the dedicated images/tutors/community folder instead of relying on a remote image URL.

The Generate full tutor button now keeps strong text and icon contrast across light, dark, and balanced themes.

Added the $299 monthly MAX plan for serious learners, with unlimited individual tutor chat and live lessons, maximum reasoning, and premium classroom model access.

Tutor classroom actions now stay in the learner-selected tab: content such as a quiz is adapted to HTML, Quiz, or the currently open surface, and only a direct navigation-only request can change tabs.

The HTML classroom tab is now called HTML consistently in tutor speech, classroom controls, help content, and technical descriptions.

Live classroom edits now carry explicit inspect-or-change intent, refresh the latest learner transcript after the preparatory read, preserve exact requests such as a 500-word expansion, retry an unchanged HTML render once, and continue the current turn without replaying the lesson greeting.

End-of-lesson tutor feedback now uses compact whole-session continuity, checkpoints sampled across the beginning, middle, and end, and the latest turns; its minimal-thinking Flash-Lite request has a smaller output and a hard latency ceiling so the recap can fall back quickly instead of waiting 30 seconds.

Lesson recaps now show staged feedback preparation with elapsed time and a progress bar, then reveal an evidence-matched reward selected from the stable VibeTutor reward set.

Account Settings now list all 97 languages documented for Gemini Live, alongside automatic matching and the English plus Urdu bilingual preference; the choice applies to text and Live tutoring while preserving explicitly requested language study.

Character, folklore, historical, and general tutors no longer refuse a request merely because it falls outside the profile subject or source story.

Broad catalog requests now receive a useful organized answer or first installment, with edition, source, date, or completeness ambiguity stated only when relevant.

When learners request a Document, Whiteboard, Canvas, code, quiz, or homework update, tutors are explicitly instructed to use the available classroom tool instead of redirecting to a specialty lesson.

Generated tutor personas were refreshed to remove competing phrases that previously told figures to keep lessons close to their own works or tales.

Cinderella now welcomes learners through hearth, ashes, hope, choice, and recognition instead of presenting resilience and workload fairness as a modern course list.

All ten story characters, ten folklore figures, and twenty historical figures now use their curated lore or method tagline as the personal introduction shown and spoken on profile pages.

The managed-content migration refreshes old generated introductions while preserving custom wording written by administrators.

Completed learner whiteboard edits now use the dedicated silent Live context channel instead of the normal speaking message path.

Structured board objects and scene details still synchronize after the learner pauses editing, while speech and chat remain quiet until the learner explicitly asks or continues the conversation.

Casual conversation is now a valid destination in text chat, voice calls, and classroom lessons, with no automatic exercise, correction, subject bridge, or prompt to get back on topic.

Tutor specialties now describe areas of strength rather than allowed-topic lists; safety, honesty, and material expertise boundaries remain active without policing benign conversation.

Each tutor keeps a stable profile-specific temperament, conversational rhythm, warmth, humor, and point of view during both teaching and casual chat without inventing a human biography.

Level filtering and level sorting now use compact labels such as Beginner, Intermediate, Advanced, Young learners, and All levels instead of hundreds of full audience sentences.

Existing tutor records were classified into a dedicated indexed database field, while profile pages retain the richer audience descriptions learners use to judge fit.

Tutor administration now separates the controlled directory level from best-fit audience copy, and new seeded or community tutors receive a compact level automatically.

Document SVG insertion now removes Markdown and XML wrappers, repairs common entity and element-closing mistakes in an inert parser, and then applies the existing strict element, attribute, URL, and size sanitization.

SVG-only markup mistakenly returned as an ordinary rich-text append is automatically rerouted to the atomic insert_svg operation instead of becoming an empty block after HTML safety checks.

Live Document SVG actions receive one bounded corrective retry after a browser validation failure, with a larger but capped generation budget and clearer compact-XML guidance.

Document changes now read the complete current rich-text state before Flash-Lite prepares an atomic revision-bound patch, preventing missing block IDs, hashes, read tokens, and exact-text errors.

Whiteboard, Canvas, Code Editor, Quiz, and Homework changes now receive a fresh destination-specific snapshot before their update tools are exposed.

Gemini Live now delegates the learner intent once while the application handles state preparation, so the voice model no longer has to copy fragile Document patch metadata itself.

Flash-Lite Document operations are validated and retried before browser execution, including safe normalization of camelCase operation names such as replaceText.

Fixed Canvas and other read-before-write actions losing the learner request between their state-read and update steps; failed Live actions now explicitly prevent the tutor from claiming an unverified classroom change.

A Canvas inspection can no longer satisfy a creation request by itself; Live and typed lessons automatically continue from read_javascript_canvas to draw_javascript_canvas without requiring the learner to ask again.

Canvas completion now verifies that the protected renderer committed a newer drawing revision, switched the classroom to Canvas, and allowed two browser paint frames before reporting success to the tutor.

If the browser cannot verify the visible Canvas update, the tutor receives an explicit failure result and cannot describe the requested drawing as though it appeared.

Explicit Canvas inspection requests now use the Canvas-specific reader instead of the generic classroom workspace or Whiteboard; stale tutor destination wording is corrected silently so the reply stays focused on the learner-requested Canvas.

Learner classroom tab changes now use a dedicated silent Live context channel: the tutor receives the active destination, while any response generated only by that state update is withheld from audio and chat.

A tutor who has already taught the learner now opens with one short greeting sentence, avoids repeating their name, role, call mode, memories, or prior lesson summary, and immediately gives the learner the floor.

Relevant plans and recent history quietly shape a returning welcome instead of becoming a recap; first recorded lessons still keep the fuller tutor introduction and input-mode orientation.

Clicking an SVG diagram now selects the whole graphic with a clear focus outline and a compact floating object toolbar.

Added 40, 60, 80, and 100 percent width presets plus a continuous 25-to-100-percent width slider that preserves the SVG aspect ratio.

Selected SVGs can be removed with Delete, Backspace, or the toolbar delete button and restored with Document Undo; validated width choices persist without saving temporary selection state.

Added six heading levels, seven text sizes, additional font choices, inline code, paragraph spacing, left-to-right and right-to-left text, and printable page breaks.

Added selection-safe find and replace plus contextual table controls for rows, columns, header cells, and table deletion.

Pasted and tutor-authored rich text now retains bounded colors, fonts, emphasis, alignment, spacing, indentation, list styles, and direction without allowing scripts, unsafe links, event handlers, or external embeds.

Added distinct tutor-voice cues for sketches, generated images, editable SVGs, board text, shapes, visual assets, Documents, Canvas drawings, code, quizzes, Homework, and supporting lesson materials.

Auto drawing mode now waits for the Classroom dispatcher to resolve the real tool before preparing speech, so the tutor says sketch, image, or SVG according to the method actually selected.

The visible classroom activity label switches to the resolved action at the same time, while speech still waits for the tutor to be quiet and stops as soon as the action completes.

Added an application message core with immutable envelopes, single-handler commands, correlation metadata, cancellation and timeout support, scoped feature routers, and bounded diagnostics.

Document, Canvas, Workspace, Learning, Quiz, Videos, and drawing skills now register the Classroom tools they own; Chat requests actions through commands instead of selecting and invoking feature controllers directly.

AI credits and authentication data now have one browser state owner with immutable snapshots, while the existing VibeTutorAuth object remains available as a stable compatibility facade.

Chat no longer reads private LiveAgent connection or playback fields, and Workspace no longer receives a Document-owned DOM status element.

Major Classroom state changes now pass through the shared validated event boundary, and Document and Workspace persistence use the resilient browser storage adapter.

Added a compact, remembered Engine selector for Vanilla Canvas 2D and p5.js 2.3.2, with responsive controls that leave the drawing surface roomy on smaller classroom screens.

Vanilla and p5.js drawings now keep separate saved layers and Undo/Redo history, so changing engines never overwrites work created with the other runtime.

The learner-selected Engine is now authoritative for Canvas reads, drawings, clears, Undo, and Redo; active Live tutors are notified immediately, and stale tutor runtime hints can no longer switch or write to the other Canvas.

Tutor-generated p5.js code runs in a short-lived protected worker with network capabilities disabled; only bounded, whitelisted drawing commands can reach the pinned p5 renderer inside its opaque, strict-CSP frame.

Fixed common flocking and motion sketches being rejected: safe local class constructors, persistent protected animation state, legacy state aliases, and p5-style vector helpers now work without exposing real browser globals or prototype access.

Both Canvas engines now syntax-check every candidate layer and complete its first protected render before changing saved work or Undo/Redo history; malformed JavaScript returns a stable syntax reference and rejected layer name while leaving the previous Canvas untouched.

Vanilla Canvas is now identified as fully animation-capable throughout Live routing, engine updates, and generation prompts, so requests such as “make this animated” read and update the current Canvas instead of being refused or redirected to Videos.

Added delayed, action-specific progress speech for images, videos, quizzes, Documents, Whiteboard visuals, Canvas work, and Homework so a longer generation no longer feels stuck.

Progress updates use the learner's selected tutor voice and a small server-cached set of fixed phrases; learner content is never sent as narration text.

Progress audio stays outside the synchronous Gemini Live tool channel, waits until the tutor is quiet, and stops before the tool result is returned so it cannot cancel an action or overlap the tutor's next response.

Fast actions remain quiet, while the existing pending activity and elapsed-time indicator continue to provide visible status and accessible announcements.

Added a remembered compact view to each high-use classroom tab, with one accessible Show controls button to restore titles, toolbars, guidance, and status details whenever they are needed.

Whiteboard tools and Document formatting now fold out of the way, while Canvas guidance and source, Quiz chrome, and Homework chrome give their interactive surfaces more room.

Tightened the classroom header, tab strip, spacing, tool rows, and editor padding for iPads, smaller laptops, short displays, and dynamic mobile viewport heights.

Kept controls keyboard accessible, horizontally scrollable when space is tight, and large enough for touch input while preserving every classroom action.

Removed automatic planned-duration announcements from learning-path and scheduled-lesson openings while retaining silent pacing context.

Reaching a planned duration now updates the call interface without injecting a spoken time-up announcement or automatic tutor recap.

Exact elapsed, total, and remaining values remain available through the authoritative timer tool whenever the learner asks.

The Live timer tool is now mechanically authorized for one use only after an explicit learner timing question; unrelated turns receive no timer data, and internal call-opening instructions can no longer be mistaken for learner timing requests.

Timer results now expire after the immediate requested answer, and Live playback stops any unrequested carry-over before it can continue, preventing tutors from appending old elapsed-time values to later teaching replies.

Added structured block reads with stable block IDs, content hashes, learner selection context, pagination, and expiring read tokens so tutors no longer guess from a truncated document preview.

Document changes now apply as one atomic patch and reject stale revisions, changed blocks, ambiguous repeated text, or invented targets without partially changing learner work.

Typed tutoring now supports a bounded read-edit-verify tool sequence, while Gemini Live carries the exact structured patch directly to the browser instead of translating it through another prose prompt.

Whole-Document replacement and clearing now require explicit learner intent, and every successful tutor edit reports a post-apply revision and document hash for verification.

Fixed incomplete append, section, heading, and exact-text operations by validating required fields early and safely recovering common model field mix-ups before HTML sanitation.

Added revision-bound SVG insertion to atomic Document patches so tutors can generate, sanitize, caption, and place accessible SVG teaching diagrams without redirecting learners to the Whiteboard.

Added an authoritative read-only Live timer tool that returns exact elapsed, planned-total, and remaining values with both minutes and seconds instead of estimates.

Microphone and typed timer questions now use the same visible call clock during the current turn, without a delayed follow-up prompt that could compete with the tutor response.

Updated the Admin System tool inventory to distinguish the common exact timer from classroom-only action delegation.

Added a proactive seven-minute connection refresh that uses the latest resumable session handle, waits for classroom tools to finish, and keeps lesson context across the replacement socket.

Long-lived, quiet 1008 lifecycle aborts can now recover even when the provider closes without first sending a GoAway warning; recent request rejections remain visible instead of being retried blindly.

Connection diagnostics now separate total lesson time from current socket time and record the refresh policy, automatic lifecycle recoveries, and latest recovery timing.

Kept sliding-window context compression enabled alongside session resumption, with clearer in-call status while a planned connection handoff is underway.

Long lessons now renew the short-lived Gemini access credential before its 30-minute expiry and resume with the latest session handle instead of reusing an expired token.

Credential renewal keeps the active socket available while it requests replacement access, retries transient failures within a bounded window, and records renewal counts, timing, and reasons in Live diagnostics.

Tutor-generated Canvas code now unwraps complete draw(ctx, width, height, time, frame) declarations before validation, storage, source display, and iframe rendering, including repeated accidental wrappers.

Dormant function-only Canvas programs now return a precise structure reference and correction guidance instead of reporting success for a blank drawing.

Fixed a code-history state variable that was accidentally scoped inside a text helper and could stop the classroom workspace script during startup.

Moved live microphone frame capture to AudioWorklet in supported browsers, while retaining a compatibility fallback for older browsers.

Added fresh classroom and chat asset versions so the corrected workspace and microphone code load immediately after deployment.

Added a Pro-only Videos workspace where learners can request 4, 6, or 8 second landscape or portrait teaching clips with Veo 3.1 Lite.

Tutors can recognize video, animation, moving-demonstration, and generated-clip requests, start the matching classroom action, and open the Videos tab automatically.

Video generation now runs as a long-running job with visible progress, clear provider errors, and automatic polling instead of holding the lesson open while the clip renders.

Completed MP4 files are copied into private account storage and served with seekable playback, so clips remain available after the temporary provider download expires.

Added server-side Pro enforcement, ownership checks, CSRF protection, rate limits, a rolling five-video daily allowance, and private media URLs.

Pricing now shows the classroom media matrix clearly and enforces the same access rules: Plus and Pro can generate classroom images, only Pro can generate videos, and Free or Basic can generate neither.

Fixed tutor-triggered video actions by using Gemini-compatible string choices for clip duration, and removed unsupported Veo Lite request fields that prevented manual generation from starting.

Moved the detailed Why choose summary and learning-goal tutor comparison from individual profiles into each profile-specific Tutor guide, with direct guide navigation and accurate profile links.

Added Nano Banana 2 whiteboard image generation for learner requests and visual explanations, using compact 512-resolution images placed at the cursor or in clean board space.

Tutors now recognize explicit image, photo, photorealistic, and Nano Banana requests as generated-image actions instead of substituting a marker sketch, refusing the request, or interpreting Nano Banana literally.

Corrected the Gemini image configuration so compact 512px whiteboard images use the model-compatible payload, and classroom failures now include concise provider field details instead of only a generic invalid-argument message.

Generated images behave like semantic whiteboard assets: tutors can move, resize, layer, and inspect them, and the browser restores them with the saved board after a reload.

Quiz setup now folds automatically after a quiz is created and can be reopened from the Quiz header, leaving more room to answer and review questions.

Expanded classroom feature controls so enabled lesson plans, practice activities, notebooks, subject labs, resources, review cards, transcripts, and SVG workspaces appear and route consistently.

Added a dedicated detailed biography section to tutor profiles, bringing each tutor's background and teaching introduction together in one readable place.

Tutor guides now show verifiable quality signals from published ratings and recorded activity, including completed lessons, conversations, average session length, and privacy-safe common goals.

Added clearer next-step actions for starting a live lesson or free text chat, with concise voice-minute and tutor-pricing guidance before learners begin.

Added a one-time welcome email after a learner creates a VibeTutor account, with direct links to tutors, profile setup, and the Help Center.

Added a selected-user email composer to Admin Users with individual delivery, automatic personalization, and no shared recipient addresses.

Recorded welcome attempts and administrator campaign outcomes so mail acceptance and failures remain auditable without interrupting signup.

Added account-linked support threads with complete message history, unread reply indicators, clear status labels, and learner follow-ups that reopen resolved requests.

Upgraded the Admin contact area into a support console where administrators can read conversations, send learner-visible replies, resolve requests, and retain separate private notes.

Added support inbox links to account navigation and the footer, plus email notifications that point learners back to the full conversation when mail delivery is available.

Kept guest contact requests email-based so administrators cannot accidentally post a reply to an inbox the sender cannot access.

Increased monthly voice time to 10 minutes on Free, 180 on Basic, 360 on Plus, and 720 on Pro.

Standardized allowance enforcement across all four tiers, so Pro now stops at its documented 720-minute monthly limit instead of continuing past the meter.

Updated pricing, features, tutor profiles, Help Center guidance, and the header credit meter to use the same plan values.

Opened saved learning paths, tutor memory, weekly planning, progress tracking, review tools, and workspace exports to signed-in Free learners; monthly voice minutes remain the primary plan limit.

Added a compact live AI usage meter in the classroom and a detailed API, model, operation, token, and cost breakdown after class.

Captured Gemini Live and generateContent usage metadata, including text, audio, image, cached input, response, and thinking tokens where reported.

Clearly labels estimates as paid Standard list-price operating costs and marks partial or duration-based fallbacks when complete token metadata is unavailable.

Added an audited Admin setting for choosing Gemini 3.5 Flash-Lite or Gemini 3.1 Flash-Lite as the primary classroom dispatcher.

Added bounded retry and model failover for transient classroom AI errors without restoring Search grounding or expanding the Live tool payload.

Applied the same recovery path to custom whiteboard drawing plans and visual board inspection.

Classroom activity history now includes a concise failure reason when an action still cannot complete.

Active classroom action labels now show a live elapsed timer and retain the total duration when each tutor action completes.

Added a technical Gemini Live section to Admin System with token, WebSocket, audio, transcript, context-compression, and session-resumption details.

Documented the direct classroom delegation boundary and all available whiteboard, lesson, workspace, quiz, and homework actions.

Clarified that the server API key stays private and that classroom actions are filtered, planned, validated, and completed before the tutor reports success.

The latest learner request now defines the active agenda across text chat, voice calls, and classroom lessons.

Tutor specialties, saved goals, lesson plans, and chat history now act as supporting context rather than mandatory destinations.

Tutors finish the current thread before proposing a topic change and ask permission at most once when a transition would genuinely help.

Tutors now receive an explicit No limit policy that prevents final-minutes hints, time-pressure language, and automatic time-based wrap-ups.

Open-ended classroom plans and workspace context no longer turn elapsed time or account credits into a lesson countdown.

Positive-duration lessons retain app-driven time-up notices and end-of-lesson recap behavior.

Added a dedicated Settings section to the React admin with a clear library-assisted or Gemini-only drawing mode.

Disabling the tutor doodle library removes reusable template and algorithm tools from tutor context while leaving manual learner inserts available.

Gemini-only mode is enforced again by the server drawing planner, including stale browser requests, and every setting change is recorded in the admin audit log.

Added a complete visual direction paragraph for face, natural skin detail, eyes, expression, hair, clothing, pose, lighting, framing, background, colors, and subject-relevant objects.

Gemini now refines the creator choices into a coherent portrait prompt without overwriting them, with a larger safe prompt budget for more distinctive Pollinations images.

The finalized portrait direction appears on community tutor profiles and remains editable with its source appearance details in the React admin.

Added a searchable React course directory with publication and built-in tutor filters, curriculum totals, learner activity, and direct public-course previews.

Added course editing for tutor assignment, learning promise, audience, duration, project structure, search metadata, chapter order, objectives, practice, checkpoints, and rich structured lesson content.

Chapter updates preserve stable IDs and block removal when learner progress exists; administrators archive courses instead of deleting account-linked learning records.

Course assignment is restricted to built-in tutors, course imagery follows the selected tutor portrait, and saved admin curriculum is protected from automatic backfill replacement.

Tutor names are now normalized and unique across built-in and community profiles, with reserved-name guidance for Gemini, one automatic collision retry, and a database unique index for concurrent requests.

The React tutor admin now filters community profiles and displays creator identity, moderation state, visibility, profile totals, and public availability at a glance.

Administrators can edit the full canonical tutor profile plus the community creation brief, appearance settings, portrait prompt, image model and seed, while inspecting generation provenance.

Private and rejected community profiles are archived and removed from public pages without deleting linked learner lessons, ratings, homework, or history.

Added a Community Tutors directory and account-owned tutor studio for defining expertise, learner fit, background, personality, and lesson energy.

Gemini Flash-Lite now builds the full tutor profile, including teaching style, specialties, persona, lesson plan, classroom tools, introduction, and sample interaction.

Added an appearance studio that creates a safe, deterministic, keyless Pollinations AI portrait URL and reuses that portrait on profiles, directories, chat, calls, lessons, and classroom surfaces.

Community profiles use the canonical tutor system, with visible provenance, account ownership, generation limits, professional adult portrait validation, and full access to favorites, ratings, scheduling, voice, and classroom tools.

Community tutors no longer receive automatically seeded public courses; any legacy community-linked course is archived non-destructively and excluded from course surfaces.

Added a permanent Homework classroom tab with shared instructions, due dates, status, checklists, learner responses, tutor feedback, and account autosave.

Added bounded tutor tools for reading, creating, and updating the selected homework during typed or voice lessons, with visible classroom activity and explicit learner-response safeguards.

Added a private Homework library and full post-class editor, linked from the dashboard, account menu, profile, and class-ended screen.

Homework records use account ownership checks and version conflicts so stale browser views cannot silently overwrite newer tutor or learner edits.

Added an AI flashcard generator with five deck sizes, optional tutor context, personal study guidance, and account-saved results.

Added a searchable flashcard library with flipping study cards, mastery tracking, deck renaming, card editing, memory notes, blank decks, and deletion.

Added tutor-specific flashcard links to every tutor profile and tutor filters to generated and classroom-created decks.

Classroom Review Cards now sync to the signed-in account, and the new Generate cards action can build and save a lesson deck from the plan, practice work, and recent transcript.

Reduced Gemini Live to one lightweight classroom delegation tool while Gemini Flash-Lite selects and prepares whiteboard, doodle, workspace, document, code, SVG, lesson, resource, and practice actions.

Live-initiated visual inspection and custom doodle planning now stay on Gemini Flash-Lite, with every selected browser action checked against the local classroom tool catalog before execution.

Expanded the one-time close-code 1007 recovery path to resume an active lesson after recent client or service activity when Gemini has supplied a valid session handle.

Microphone and screen frames now use the shared validated Live message sender, so diagnostics identify the true latest outbound payload instead of an older tool response or whiteboard update.

Expanded custom Gemini drawings to five bounded detail levels from 24 to 96 paths plus an Unlimited option. Unlimited uses the maximum 48k response budget without browser or whiteboard stroke-count truncation, while preserving up to 64 exact points per path from preview through playback, Notebook, and saving.

Fixed classroom Notebook LaTeX previews so explanatory prose can be combined with inline or display formulas using dollar, parenthesis, or bracket delimiters, while direct formula-only blocks continue to render as display math.

Added a Drawing AI selector above the whiteboard for choosing fast Gemini Flash-Lite plans or higher-detail Gemini Flash plans when included with the learner plan.

Added an AI Draw toolbar action with a two-step Generate and Draw workflow, including an optional safe SVG preview built from the exact validated Gemini paths before live whiteboard animation begins.

Successfully generated AI doodles are now saved automatically as sanitized SVG blocks in the classroom Notebook for later review, editing, or deletion.

Learner-generated AI doodles no longer create chat activity entries, while tutor-triggered sketch activity remains visible in the conversation.

Added five object tabs to Doodle Studio with 19 live-sketched templates for common technology, school, everyday, nature, and transport drawings, including a laptop, mountains, and a car.

Added five selectable custom-drawing detail budgets from 8 to 48 validated paths, with up to 32 points per path, for faster sketches or richer recognizable detail.

Custom Gemini doodles now follow the generated points in exact order without automatic curve fitting or smoothing; curve algorithms remain available only for explicit algorithm lessons.

Changed the Pro voice allowance into an included-minute meter: it still reaches zero as a visual usage cue, but Pro members can keep starting and continuing calls.

Made completed tutor doodles select as one group with rotation, proportional corner scaling, independent width and height handles, and direct W/H fields.

Kept the same handwriting font on doodle labels after the live drawing animation is committed to the whiteboard.

Hardened voice lessons against rejected classroom text updates by repairing malformed browser text, identifying each update source in diagnostics, and automatically resuming once when Gemini provides a usable session handle.

Lesson recaps now choose a varied reward that stays stable on refresh and avoids immediately repeating the learner’s previous reward.

Lesson-end screens now hold one calm anticipation state, then reveal the final tutor review and stable reward together after at least one second instead of replacing placeholder feedback.

Added a prominent first-person introduction to every published tutor profile, with a tailored introduction for Amara and profile-aware defaults across the directory.

Connected each introduction to chat and first-call context so the tutor learners meet in a lesson matches the personality presented on the profile.

Added introduction editing to the tutor admin workflow while preserving administrator-written text through later seed updates.

Added a visible tutor-fit comparison with best-for goals, strengths, specialties, teaching approach, an example lesson, and the learners who benefit most.

Added editable example conversations that demonstrate each tutor's correction style, explanation depth, and learner participation before a lesson begins.

Added transparent tutor quality signals for learner satisfaction, completed lessons, conversations, average session length, and privacy-safe common learning goals.

Strengthened tutor profile calls to action with free text-lesson prompts, real five-minute call presets, and a clear next-step panel after the detailed profile.

Added a personalized first-lesson preview to every tutor profile, with subject-aware phases, preparation guidance, and a flexible lesson expectation.

Added transparent Week 1, Week 2, and Week 4 tutor progress roadmaps with subject-aware targets, outcome disclaimers, and an action to adapt the roadmap in chat.

Added compact tutor comparisons that place the current profile beside relevant published alternatives using real specialties, teaching focus, and direct profile links.

Strengthened tutor personality with a memorable, editable signature learning experience on every profile and carried it into text-chat and live-call behavior.

Added tutor-aware profile FAQs covering beginner fit, grammar correction, lesson memory, speaking exercises, and writing practice, with matching visible FAQ structured data.

Added a rich, tutor-specific character persona to every profile, carried it into text and live lessons, and made the complete behavior prompt editable in admin without overwriting custom versions.

Clarified the live-call start dialog with an explicit input-choice instruction, stronger green start actions, directional cues, and clearer optional classroom wording.

Made tutor conversations more immersive and relatable with profile-fitting natural speech, specific emotional responsiveness, fewer generic assistant phrases, contextual identity disclosure, and direct honest answers whenever a learner asks whether a tutor is AI.

Focused the shared header by moving Courses to footer navigation and added an always-visible, theme-aware voice-credit meter with live remaining-minute updates and a clear Free-plan upgrade path.

Put the unused side of public breadcrumb bars to work with the live tutor-directory count, Free text-chat availability, and a direct Browse action.

Added private, revocable lesson-calendar subscriptions for Google Calendar, Apple Calendar, Outlook, and other iCalendar-compatible apps, with read-only event details and automatic schedule updates on provider refresh.

Added an extensible classroom tutor-skill registry and progressive doodle studio with reusable educational templates, tutor freehand sketches, narrated drawing cues, playback controls, reduced-motion support, editable layer persistence, and one-step undo.

Added algorithmic tutor drawing with Bresenham lines, midpoint circles and ellipses, De Casteljau Bezier curves, Chaikin smoothing, regular and star polygons, parametric spirals, and L-system trees.

Added original concept doodles: tutors can ask Gemini Flash Lite for SVG-style line art of objects that are not in the library, then sketch the validated paths live with the animated marker in typed or voice lessons.

Added profile-specific guide pages built from each tutor's subjects, learner level, personality, specialties, lesson rhythm, custom tools, and responsible-use boundary.

Added tailored workflows for text chat, voice calls, Whiteboard, Practice, Notebook, Document, adaptive Subject, Pronunciation, or Code Lab, Review Cards, Transcript, and SVG Studio where relevant.

Added a paginated tutor-guide directory, direct guide actions on tutor profiles, breadcrumbs, Article and LearningResource data, tutor portraits, and complete sitemap coverage.

Kept the existing guide to choosing an online AI tutor as the general comparison resource and linked it alongside the new tutor-by-tutor playbooks.

Expanded the catalog to 260 topics across 26 balanced categories spanning education, work, business, technology, health, finance, practical life, media, entertainment, games, sports, travel, hobbies, sustainability, and community.

Each topic page now brings together learning outcomes, tutor prompts, matched AI tutors, related courses, assessments, revision resources, and nearby topics.

Related tutors now appear directly after each topic overview, with topic-aware chat and voice lesson actions.

Topics are available from the footer and site search, with breadcrumbs, structured data, and complete sitemap coverage.

The main directory now uses linked category tiles with visible topic counts, responsive layouts, search filters, and 24-topic pagination.

Added indexable category landing pages, self-canonical pagination, richer topic breadcrumbs and learning-resource metadata, plus clearer paths from organic topic visits into free tutor chats and learner profiles.

Reorganized help around tutor discovery, text and voice lesson modes, scheduling, learning paths, collaborative classrooms, progress, account data, plans, and troubleshooting.

Added current guidance for calendar rescheduling, standalone lessons, classroom sharing, rich documents, code and SVG tools, self-tests, certificates, messages, themes, and plan changes.

Expanded visible FAQs and matching structured data for hosted AI access, sign-in requirements, live-call errors, classroom sharing, paid learning paths, and scheduled downgrades.

Expanded the guide with practical criteria for subject fit, learner level, language, tutor personality, lesson format, tools, and responsible AI use.

Added current interface screenshots for tutor discovery, full tutor profiles, and live lesson options.

Added an on-page contents menu, comparison tables, a first-task framework, continuity guidance, a decision checklist, and direct next steps.

Added a Certificates earned section with course completions and the latest saved result from each subject self-test.

Each learning record includes its earned date, result, issuer, stable reference, and a link back to the source course or assessment.

Kept certificate status derived from saved course progress and test attempts so learner records remain consistent with their source data.

Added a call-duration selector to direct tutor calls, with No limit as the default and choices from 15 minutes through 3 hours.

Scheduled lessons and learning-path sessions continue to use their saved duration without asking the learner to choose it again.

Clarified that an open-ended lesson timer does not bypass the voice minutes available on the learner account.

Added a Calendar tab to Lessons with local-time day and week views, tutor portraits, lesson types, and direct lesson editing.

Learners can drag lessons to a new time, resize them to change duration, or use the exact date, time, and duration editor.

Linked signed-in dashboard and profile shortcuts directly to the Lessons calendar view.

Added profile-specific personality traits and classroom-presence descriptions to tutor pages and AI lesson context.

First live sessions now include a formal personal introduction and call-mode guidance, while returning sessions resume with a shorter welcome.

Classroom tool activity now appears in chat history while tutors inspect or update whiteboards, documents, practice, code, and other lesson areas.

Added a dedicated minimize control that collapses chat to its tutor title bar and restores it without losing the conversation.

Minimized live calls retain the lesson timer, mute and end controls, connection indicator, and animated voice activity.

Classroom whiteboards and other workspaces now reclaim the chat column while the tutor window is minimized.

Added persistent lesson scheduling with local date, time, and timezone support from both the Lessons page and full learning-path session details.

Tutor profiles now include Schedule lesson beside Chat and Call, with an open-lesson picker when a tutor has multiple available sessions.

Scheduled lessons can be rescheduled or returned to an unscheduled state without changing their learning-path content.

The weekly planner now creates timed appointments from a preferred start time while preserving dates and times the learner scheduled manually.

Added a rich text Document tab with headings, fonts, text sizing, emphasis, colors, alignment, lists, links, indentation, undo and redo, clear formatting, copy, HTML download, local autosave, and tutor collaboration tools.

Document selections now remain visible and recoverable while using toolbar controls, dropdowns, color pickers, and inline link or table panels.

Expanded document tools with superscript, subscript, text case, line spacing, checklists, structured tables, date insertion, unlink, and select all.

Merged the Code Lab editor and syntax preview into one scroll-synchronized, highlighted editor and anchored the More workspace menu directly below its button.

Added responsive line numbers that remain aligned with the highlighted source while editing and scrolling.

Moved learner details, tutor context notes, and theme selection to a dedicated authenticated Settings page linked from the account menu.

Added a compact Settings shortcut beside the header account control, with a full-width labeled version in mobile navigation.

Added responsive tables, headings, lists, task lists, blockquotes, links, dividers, strikethrough, and sanitized rich formatting to tutor chat while retaining syntax-highlighted code blocks.

Added automatic GoAway handling and session resumption so the browser refreshes the Live connection before its roughly 10-minute connection limit instead of ending the lesson with a policy error.

Turned the compact header account control into an accessible dropdown for Profile, Settings, Messages, Lessons, Favorites, plan management, Admin access when available, and Sign out.

Replaced the multi-panel learner workspace with a clearer Lessons page for upcoming sessions, recent past lessons, voice allowance, and saved learning paths.

Weekly schedules now keep machine-readable lesson dates, ignore expired schedule entries, and fall back to the next unfinished session in each active path.

Renamed learner navigation and account shortcuts from Workspace to Lessons while preserving old workspace links through a compatibility redirect.

Classroom tutors can now inspect an on-demand rendered whiteboard view alongside structured notes and diagram geometry instead of relying on text-note metadata alone.

Improved classroom drawing with smoother pen strokes, true erasing, learner arrow and rectangle tools, and explicit local-save and tutor-sync states.

Added a searchable classroom library with premade shapes, subject icons, emoji, and reusable SVG teaching diagrams for flow charts, graphs, cycles, timelines, biology, and more.

Whiteboard visuals can now be selected, moved, resized, rotated, duplicated, recolored, deleted, and restored through operation-aware undo and redo; tutors can insert and identify the same semantic assets.

Expanded the classroom insert library with additional teaching shapes, subject icons, feedback emoji, charts, comparison tools, science visuals, and planning diagrams; inserted visuals now support direct corner-drag resizing.

Leaving a classroom now uses a focused confirmation step that safely stops any live call, records used time, and saves the whiteboard before navigation.

Added a responsive class recap with live lesson duration, planned-time progress, board activity, tutor-sync state, completion time, and useful next actions.

Expanded collaborative classrooms with persistent Rich Notes, a shared highlighted Code Lab, and a safe SVG Studio; live tutors can read, open, add, and revise these lesson surfaces through dedicated tools.

Reorganized classroom navigation around Plan, Whiteboard, Practice, Notebook, and an adaptive subject lab, with Resources, Review Cards, Transcript, Code Lab, and SVG Studio grouped under More; recap now appears near the planned lesson end.

Classrooms now open directly on the whiteboard while a concise AI lesson plan is generated automatically from the tutor, lesson request, duration, and linked learning-path session.

Live call failures now show a safe diagnostic reference, elapsed time, connection phase, close code and meaning, service state, recovery advice, and the exact point at which the call stopped.

Added local date-and-time labels with seconds to tutor chat messages, with the full timezone-aware timestamp available on hover.

Moved Gemini 3.1 Live calls to the current v1beta ephemeral-token endpoint and added strict JSON-safe validation for classroom function results before they are returned to the tutor.

Prevented classroom voice calls from failing when microphone input overlaps a tutor whiteboard or workspace action; cancelled tool calls are now handled safely and typed input waits briefly for the action to finish.

Replaced the overly aggressive Live context-compression limit with Gemini 3.1 adaptive compression, aligned ephemeral-token provisioning with v1beta, and added safe microphone, token-usage, and session-state diagnostics.

Added full-screen tutor classrooms with a shared whiteboard, learner drawing and text tools, local autosave, image export, and AI tutor diagram controls.

Added a Windows-style True dark theme with near-black pages, charcoal surfaces, gray borders, and a restrained blue accent.

Applied every saved appearance theme to the React admin console, including tables, reports, charts, forms, dialogs, and status messages.

Introduced this public changelog so learners can review meaningful product improvements over time.

Added a compact lesson-history shortcut to the signed-in header and aligned lesson-record wording across learner pages.

Added a protected admin application map with live runtime facts, architecture diagrams, dataflow documentation, schema inventory, and a complete feature catalog.

Simplified pricing around tutor access, hosted voice minutes, learning paths, and monthly or discounted yearly billing.

Made tutor cards more compact and consistent, with aligned actions, image-overlay bookmarks, pagination, filtering, and sorting.

Added tutor lesson pricing to profile summaries and call actions so learners can see the lesson context before starting.

Moved Gemini access behind the VibeTutor server so learners no longer need to provide a personal API key.

Grouped tutor-specific text and voice settings, with clearer controls for pace, correction style, reply depth, lesson mode, and voice.

Voice calls now stop immediately after a connection error, record the problem in chat, and stop consuming minutes at the failure point.

Added a dedicated message center that keeps each signed-in learner’s tutor conversation history together.

Expanded self-tests across the subjects covered by the site and made saved results available as relevant tutor context.

Tutors now greet learners by name and can use saved learner notes and recent conversation context for more continuous lessons.

Added a React admin system for users, tutors, reviews, calls, learning paths, contacts, audit history, and system readiness.

Introduced compact traffic, usage, engagement, and system reports with realtime visitor activity and bot separation.

Improved country reporting with CDN flags, country-only resolution, short retention, and privacy-conscious visitor identifiers.

Added richer courses with interactive examples, SVG diagrams, mathematical notation, flow charts, and highlighted code where appropriate.

Introduced public revision pages, flashcards, quizzes, and subject-matched tutor recommendations.

Strengthened public-page breadcrumbs, image metadata, sitemap dates, course disclosures, and search-friendly page structure.