Skip to main content
Support

AssisT Changelog

Every version, every change — transparent by design.

AssisT follows Keep a Changelog format and Semantic Versioning. Full release notes and downloadable ZIP files are on GitHub Releases. Source code and history are public on github.com/Fiavaion/AssisT.

v1.0.0 Latest Chrome Web Store release

AssisT 1.0.0 - First stable release

Everything here has landed since 0.9.3, including the 0.10.0 work that was prepared but never published. Full notes and the download go up on GitHub Releases when the release is published.

AssisT 1.0.0 goes out through the Chrome Web Store first. The tag on the public repository follows the Chrome Web Store release rather than leading it, so the newest tag on GitHub may be a version behind what is described here.

New

  • Nothing of yours leaves your device until you say yes. AssisT asks before it sends anything you select, type or say to an outside service, one question per service: the dictionary, translation, Translate Page, checking a citation online, downloading a Read from Images language other than English, dictation, lecture capture and each cloud AI provider. On a course page the question opens in a small AssisT window, so Chrome can ask for its permission in the same press. Each question names exactly what would be sent and remembers your answer. First-run setup asks only about the services your own choices would use.
  • A Your data tab in Advanced options lists every service with your answer and every saved AI key, with a Remove button and a way to change your mind at any time. A change takes effect on the very next request.
  • Mind Map (beta) - an outline-first planning and revision tool that works with AI switched off. The outline is a WAI-ARIA tree you can drive entirely from the keyboard, and the canvas is a second view of the same map. Six templates, a searchable library, a File menu, export to Markdown, OPML, JSON, SVG, PNG and FreeMind, read aloud, study mode, "Add to mind map" from any page, and "Map this page" from a page's headings. Alt+Shift+M opens it.
  • Mind maps move between AssisT and other tools. Open a map made in XMind, Coggle, MindMeister, Freeplane or FreeMind, or a plain indented outline, and save one as a FreeMind .mm file that all of them open. Files are read on your own computer, and a file you open always becomes a new map.
  • Work on one branch of a mind map at a time with Focus (Alt+Shift+F), select several nodes and act on them together with one undo, drag a node above, below or inside another, and move a node without dragging with "Move to…" (Alt+M).
  • The new popup is the only popup: every support on one screen, grouped as Reading, Writing, Seeing, Focus and Study, with a light and dark theme. Search it by what you want to do, such as "read to me" or "too bright", as well as by name. Matching happens on your device.
  • The citation manager, rebuilt: in-text citations with page and paragraph locators, a checker that compares your essay against your reference list on your device, a restorable trash, bulk select, book chapters and theses, undated sources, Zotero BibTeX and RIS import, add by DOI, and a style that stays where you set it. Harvard, APA 7, MLA 9 and Chicago 17.
  • Maths Support - an on-page calculator with a scratchpad, and large-number formatting that groups digits, keeps the value copy-safe and speaks the number in words to a screen reader. Off by default.
  • Companion features, each with its own switch: study-tools discovery, Personalised AI, equation read-aloud, Authorship Shield (a private local writing history you can export), flashcards with spaced repetition, a reading-level setting for AI summaries and simplification, a lecture companion with live captions and notes, and hands-free voice commands.
  • See what the learning record holds about you, and delete it in one step. "Keep a record of how I work" and "Offer help when reading gets hard" are off until you turn them on, and are described in plain words first.
  • Make a flashcard from the highlight menu on any page.
  • New shortcuts: Alt+Shift+V starts and stops dictation from anywhere on the page, and Alt+Shift+U moves focus to Undo in the Saved tray. The mind map's own shortcuts can be changed and checked for conflicts like every other AssisT shortcut.
  • Dictation says in plain words when it is starting, recording, paused, or blocked because the microphone is not allowed.
  • AssisT's own controls declare their purpose with WAI-Adapt attributes, for assistive tools that show symbols instead of words.
  • An optional OCR accuracy boost for low-quality scans and small text, higher-fidelity OCR layout, and equations read aloud through the Speech Rule Engine.
  • Read from Images works in English with nothing downloaded: the recognition engine and the English language file ship inside AssisT. The file for another language is downloaded once, after you allow it, checked, and kept on your device.
  • The code that runs a Browser AI model ships inside AssisT. Only the model itself downloads, from Hugging Face, when you choose one.
  • A configurable local-AI server address in the Local AI panel. The public build accepts loopback addresses only.
  • A welcome page that opens on install, with AI setup and "Discover your tools", and a "Report a Bug" link in the popup.

Changed

  • Read Aloud now starts with a voice that runs on your device. Online voices are still available, and the voice picker says which company each one sends your text to.
  • AssisT no longer asks at install for access to any outside service except Hugging Face for Browser AI models: not the dictionary, MyMemory, Crossref, jsDelivr, Google Gemini, Anthropic, OpenAI, Perplexity, Semantic Scholar or a local AI server. Chrome asks when you choose one, and the popup offers an "Allow access" button if it is needed.
  • The source check on a citation shows a plain rating (Looks reliable, Check before you use it, Hard to tell) with its reasons, instead of a numeric score.
  • Syllable colours is the default dyslexia reading mode.
  • Very plain summaries are short and in everyday words, and simplified text keeps the proper term and explains it in brackets. Neither starts with a line such as "Here is a simplified version".
  • References follow each style's published rules more closely: editions, page ranges, organisation authors, authors who share a surname, undated works and social media posts.
  • Saved keyboard shortcut sets pick up new defaults instead of ignoring them.
  • The Local AI panel lists only models that can hold a conversation, and local failures say what actually went wrong.

Fixed

  • A setting you change in the popup no longer overwrites one you changed on the page a moment earlier, and font and spacing settings are remembered again.
  • Pages already open when AssisT is installed, or when you allow it everywhere, get AssisT straight away.
  • Canvas, Moodle and Google Classroom integrations load again, announce themselves only on their own pages, and Read Aloud works on Moodle and Google Classroom as well as Canvas.
  • Dictation works in Canvas's rich content editor again.
  • A note you add to a paragraph you selected by triple-clicking is kept.
  • Keyboard and focus: popup buttons, the citation manager, the AI setup page, the magnifier, the reading guide, the study timer, Speed Read and every AI panel can be used from the keyboard, Escape closes them, and focus returns to what opened them.
  • Side panels fit a 320px-wide screen, and on-page controls share one dock instead of covering each other.
  • Grammar colours tag every word correctly, and several colours and badges now meet WCAG AA contrast.
  • Screen Color Overlay and Custom Cursor work in Google Docs and Google Sheets.
  • Local reasoning models no longer leak their thinking into answers, and the on-device AI picks a model that is actually installed.
  • Removing duplicate citations keeps every quote, page, note and tag, and RIS export no longer breaks a note across two lines.
  • The discovery quiz switches on every feature it recommends.

Removed

  • The DeepL and Microsoft Translator options, which nothing in AssisT could call. Any key saved for them is deleted. MyMemory is the translation service.
  • The Azure Speech option, which nothing could reach.
  • The "Try the new AssisT" switch, the Layout mode option and the stuck-reader rescue prompt, which "Offer help when reading gets hard" replaces.
  • The field for pasting an institution branding key. Branding set by institution policy, and branding already applied, keep working.

Security and privacy

  • Every request to an outside service goes through one place in the extension, which checks your answer first. A missing or unreadable answer is never treated as a yes.
  • A settings file you export never carries an AI key or a consent answer.
  • The console no longer records what you dictate, what an AI feature answers about your work, or what is read aloud.
  • Saved keys are stored encrypted, and the hidden page that holds your citation library accepts messages only from AssisT's own background script.
v0.9.3 GitHub

AssisT v0.9.3 — Citation Manager overhaul + annotation fix + Firefox scaffolding

What's changed

Added

  • Citation Manager overhaul — full rebuild of the citation feature:

- citation-types.js keystone module with 12 source types and typed field definitions.

- Four citation styles: APA 7th, MLA 9th, Chicago 17th, Harvard — each in its own dedicated module.

- One-click save tray — floating panel for quick project assignment without opening the full manager.

- AI citation analyser — optional AI pass evaluating source quality, relevance, and academic rigour.

- Focus trap (citation-focus-trap.js) — WCAG 2.1 compliant keyboard containment for all citation modals.

- Highwire citation_* meta-tag extraction with non-empty title fallback chain.

- 210 unit + E2E tests covering all formatters, storage, save tray, AI analyser, and UI.

  • Cross-browser API abstraction (src/core/browser-api.js) — webextension-polyfill wrapper for browser.* API across Chrome, Firefox, and Safari.
  • @fiavaion/lms-adapter package scaffold — standalone LMS detection + course/assignment context (Canvas, Moodle, Classroom). NLnet deliverable.
  • @fiavaion/local-ai-router package scaffold — four-mode local AI routing library (Ollama → WebLLM → Chrome Prompt API → cloud). NLnet deliverable.
  • Pluggable prompt registry (src/ai/prompts/registry.js) — prompts keyed by feature × discipline × user profile. NLnet deliverable.
  • Firefox build target — BROWSER=firefox npm run build:firefox → .vite-firefox/ with __IS_FIREFOX__ build flag.

Fixed

  • DOMPurify name-clobber — edit modal silently dropped citation title on save; fixed by reading textContent after sanitisation.
  • Gemini AI model — replaced retired gemini-1.5-flash (404) with gemini-2.5-flash; fixed 429 retry loop.
  • Annotation highlight race condition — Range invalidated after await; span now inserted synchronously before storage write.
  • Organize Mode drag & drop — SortableJS now targets .accordion-section; drag handles were reaching the wrong ancestor.
  • Organize Mode visibility toggles — accordion header mousedown no longer suppresses toggle clicks; restored on exit.
  • Double-fire on OrganizeMode controls — removed duplicate click + mousedown registration.
  • STT Beta label removed — Speech-to-Text graduated from beta.
  • service-worker — chrome.runtime.id → browser.runtime.id via abstraction layer.
  • Citation Manager marked Alpha — labelled in popup to set expectations while UI continues to evolve.

CI

  • GitHub release now triggers Cloudflare Pages rebuild of the Fiavaion website.
  • deploy.yml ZIP source dir corrected (Output/ → .vite/).
  • notify-website.yml opts into Node 24 ahead of GitHub's June 2 forced Node 20 deprecation.

---

Full changelog: https://github.com/Fiavaion/AssisT/blob/main/CHANGELOG.md

Chrome Web Store: https://chromewebstore.google.com/detail/assist-adaptive-augmentat/dkekfjomoacmhbkekjkngmpbdlljjfhi

v0.9.2 GitHub

v0.9.2 - AI Setup Improvements

Ollama model catalog with 11 curated models organised by category (General, Reasoning, Multilingual, Code, Vision) and hardware tier (4GB/8GB RAM). WebLLM cache-clear button to free downloaded model storage. Honest Browser AI speed warning (30-90s per response) added to setup wizard. Full CHANGELOG.md now in repository following Keep a Changelog and semver standards.

v0.9.1 GitHub

v0.9.1 - GAAD 2026 Public Launch

First public release, launched on Global Accessibility Awareness Day (21 May 2026). Awarded the Chrome Web Store Featured badge. Fixes: WebLLM offscreen document for WebGPU access, Discovery Quiz storage paths, localhost:11434 host permission for Ollama. Install from the Chrome Web Store: https://chromewebstore.google.com/detail/dkekfjomoacmhbkekjkngmpbdlljjfhi

v0.9.0 GitHub

v0.9.0 - Final Beta

Pre-launch hardening before GAAD launch. See CHANGELOG.md for full details.

v0.1.2 GitHub

v0.1.2 - AI Overhaul & Multi-Provider Expansion

CWS resubmission — Phase 2 AI overhaul and multi-provider expansion

Chrome Web Store resubmission following a complete overhaul of the AI layer: Browser AI (WebLLM), a multi-provider cloud AI router, a guided AI Setup Wizard, and a shared routing client used by all 8 AI features.

Added

  • Browser AI (WebLLM) — AI models run entirely in the browser via WebGPU; zero installation required. 8 models available (0.6 GB–6 GB).
  • Inline AI mode switcher in popup — switch Off / Cloud / Browser AI / Local AI with one click, no wizard redirect needed.
  • AI Setup Wizard — step-by-step guided setup for all AI modes with hardware detection and model recommendations.
  • Model registry (model-registry.js) — centralised catalogue with per-task routing defaults.
  • Multi-provider cloud AI routing — Anthropic (Claude), OpenAI (GPT), Google (Gemini), and Perplexity all supported from a single provider selector with dynamic model lists.
  • Persistent AI status bars on all 8 AI feature panels — shows current mode, availability, and last result at a glance.
  • Shared AI routing client (ai-feature-client.js) — single source of truth for mode detection, availability checking, and generation; eliminates per-feature getCurrentModel() inconsistencies.
  • Multi-Document Compare — analyse and contrast multiple text sources side-by-side.
  • Google Docs Speech-to-Text — microphone button now appears inside Google Docs' custom rich editor.
  • Text customisation sync — push font/spacing settings to all open tabs simultaneously (opt-in toggle).
  • CI pipeline — GitHub Actions with caching and redundant-run cancellation.
  • Default Ollama model changed to qwen3:8b-q4_K_M for superior JSON compliance and instruction following.
  • Z-index scale system (src/utils/z-index.js) prevents UI stacking conflicts with LMS page elements.

Changed

  • All 8 AI features migrated to shared routing client.
  • Local AI model picker redesigned with live Ollama detection and dropdown.

Fixed

  • Knowledge graph local AI pipeline now works on first run without reload.
  • API key encryption — keys now encrypted with AES-256-GCM in secure-key-storage.js; previously stored as plain text.
  • STT microphone button positioning — was off-screen on some Canvas layouts.
  • STT enabled on Google Classroom — fixed iframe injection and storage mismatch.
  • Multi-document compare drag rewritten with Pointer Events API and setPointerCapture — eliminates listener leaks and unreliable drag.
  • Text customisation sync storage race condition — now uses chrome.tabs.sendMessage instead of shared chrome.storage.local.
  • Canvas white-text inheritance — added color: #333 to AI panel roots so text is readable on dark-themed Canvas pages.
  • SPA navigation pre-audit: all async handlers given explicit timeouts; z-index conflicts resolved across Canvas, Moodle, and Classroom.
v0.1.1 GitHub

v0.1.1 - Phase 2 Feature Expansion

Phase 2 — 11 new features built on the TTS/STT foundation

First CWS submission attempt. Major feature expansion round adding AI-powered academic tools, citation management, knowledge graphs, and a full user profiles system.

Added

  • Citation Management — Harvard Style (Cite Them Right, 13th ed.) generator for websites, books, journals, videos, and social media. Includes bibliography manager, CRAAP test evaluation, and BibTeX/RIS/JSON/CSV export.
  • Online citation verification via CrossRef and Semantic Scholar APIs.
  • Knowledge Graph — interactive concept-connection visualisation from selected text.
  • Socratic Tutor — guided learning through AI questions rather than direct answers.
  • Study Path Generator — personalised learning sequences for any topic.
  • Assignment Breakdown — decomposes assignment instructions into step-by-step checklists.
  • Text Simplification — three complexity levels; trailing AI conversational artifacts stripped automatically.
  • Summarization — brief, detailed, bullet points, or ELI5 modes.
  • Translation — 35+ languages via MyMemory (default) and LibreTranslate (fallback); rate-limited.
  • Highlight Menu — contextual toolbar on text selection with Read Aloud, Dictionary, Translate, AI Summarize, and more.
  • Text Statistics — word count, characters, sentences, reading time (Ctrl+Shift+W).
  • Discovery Quiz — recommends features based on learning style and needs.
  • User Profiles system — save, load, export, and import accessibility configurations. 9 neurodivergent profiles (ADHD, Autism, Dyslexia, Anxiety, Motor, Low Vision, Cognitive Overload, Sensory, General Learning).
  • 7 Speech-to-Text profiles with customised timing, UI, and feedback.
  • Drag-and-drop section reordering in popup (Organise Mode).
  • Keyboard shortcut presets — Default, Minimal, One-Handed.
  • Feature maturity badges (Alpha, Beta, Experimental) displayed in popup.
  • Optional <all_urls> permission flow — single-click "Enable Everywhere" for non-LMS sites.
  • DOMPurify XSS protection on all user-facing inputs.
  • Gemini Nano (Chrome built-in AI via Prompt API) added as a fourth AI mode.

Removed

  • Stargardt Vision Mode removed — 11 experimental files deleted (eye tracking, APVUI rendering, scotoma profiling, calibration wizard). Feature had low adoption and introduced significant maintenance overhead.

Fixed

  • Content script injection for non-LMS sites when <all_urls> granted.
  • web_accessible_resources updated for Vite-bundled asset paths.
  • Various keyboard shortcut layout conflicts.

Security

  • Optional permissions model — <all_urls> is user-granted on demand, not requested upfront.
  • GDPR compliance section added to privacy policy.
v0.1.0 GitHub

v0.1.0 - Initial Release

Initial release — TTS/STT foundation with Canvas LMS integration

First public version of AssisT, establishing the core text-to-speech and speech-to-text foundation with full Canvas LMS integration.

Added

  • Text-to-Speech with synchronised word-by-word highlighting. Customisable rate (0.5×–2.0×), pitch, volume, and 8 highlight colours.
  • Speech-to-Text using Web Speech API. 60+ voice commands for editing, formatting, and navigation. Smart auto-punctuation (Automatic / Assisted / Manual). Custom vocabulary with domain presets. Confidence feedback with colour-coded accuracy indicators.
  • Reading Mode — distraction-free view powered by Mozilla Readability with full TTS integration (Alt+R).
  • Dyslexia-Optimised Reading Modes — Bionic Reading, Syllable Colours, Grammar Colours (Alt+Y).
  • OCR & Screenshot Tool — extract text from images and PDFs using Tesseract.js offline (100+ languages, Alt+O).
  • Focus Mode — dims surrounding content, centres reading area (Alt+F).
  • Reading Guide — horizontal cursor-tracking line (Alt+G).
  • Screen Colour Overlays — 9 comfort filters (Alt+Shift+O).
  • Reading Progress Bar — scroll position indicator (top or bottom).
  • Text Customisation — 11 fonts including OpenDyslexic, Lexend, Atkinson Hyperlegible. Line height, letter, word, and paragraph spacing. WCAG 2.2 SC 1.4.12 compliant.
  • Reduced Motion — disables all animations; respects prefers-reduced-motion (WCAG 2.1 SC 2.3.3).
  • Custom Cursor & Magnifying Lens — multiple cursor styles; floating 1.5×–4× magnification window.
  • Dictionary Lookup — inline definitions via Free Dictionary API (Alt+Shift+D).
  • Sticky Notes & Annotations — per-URL floating notes in IndexedDB with tag/export/sidebar (Alt+N).
  • Pomodoro Timer — configurable work/break intervals with ADHD-optimised defaults.
  • Media Control — blocks autoplay video and audio.
  • Canvas LMS integration — content script auto-injected on *.instructure.com / *.canvas.com; Canvas Quiz Helper with TTS and keyboard navigation.
  • Moodle and Google Classroom host permission support.
  • 14 customisable keyboard shortcuts with Chrome-conflict detection.
  • Manifest V3 architecture with Vite + @crxjs/vite-plugin build pipeline.
  • Jest unit testing (979 tests) + Playwright E2E infrastructure.
  • WCAG 2.2 AA compliance throughout.

Development History

These internal checkpoints were created during the sprint-based pre-release phase (October–November 2025). They are preserved in the git repository for history and rollback reference but are not versioned releases, and they are not browsable on the public repository — the tag names are listed here for reference only.

Tag Milestone
MVP-TTS-Stable-v1.0 Stable TTS milestone — word-highlight sync confirmed working
Sprint8-Testing-Complete Playwright E2E infrastructure scaffolded
Sprint7-Canvas-Profiles-Complete User Profiles system — 9 neurodivergent presets
Sprint6-ReadPage-Simple-v1.0 Read Entire Page feature
Sprint5-STT-Complete-v1.0 Speech-to-Text with Web Speech API (60+ voice commands)
Sprint4-Canvas-Foundation-v1.0 Canvas LMS content-script integration
Sprint3-Complete-v1.0 Focus Mode, Text Customization, Reading Guide
Sprint2-Foundation-v1.0 Initial popup and TTS controller