AccessBridge
Accessibility that notices.
A Chrome extension that silently detects user struggles and auto-adapts any web app in real time. Zero disclosure. Zero configuration. Powered by on-device AI.
Get AccessBridge for Chrome
Runs entirely on-device. No server required. No data leaves your browser.
Download Extension (.zip) v…Download & Unzip
Download the extension package and extract to any folder
Enable Developer Mode
Open chrome://extensions and toggle Developer Mode on
Load Extension
Click "Load unpacked" and select the unzipped folder
How It Works
Three steps to ambient, invisible accessibility
Install Extension
Download the AccessBridge Chrome extension and load it in developer mode. One-time setup, under a minute.
Browse Normally
Use any website as usual. AccessBridge runs silently in the background, analyzing behavioral signals without interrupting your workflow.
Auto-Adapts
When struggle is detected, the page automatically adapts: font size, contrast, layout, focus mode, and more adjust in real time.
See It In Action
AccessBridge detects struggles and responds instantly, without any user configuration
Real-time behavioral signal monitoring
Ambient Intelligence at Work
AccessBridge continuously monitors subtle behavioral signals to understand user needs. When struggle patterns emerge, the AI decision engine triggers targeted adaptations without requiring any explicit action.
- Detects motor impairment via click accuracy and cursor tremor
- Identifies vision struggles from scroll and zoom behavior
- Adapts layout complexity based on cognitive load signals
- All processing runs locally -- no data leaves the browser
Core Features
Ambient, invisible accessibility that adapts in real time to user struggles
Struggle Detection
- Watches how you interact — scroll speed, click accuracy, typing rhythm
- Spots frustration patterns in seconds, not minutes
- Updates a live "struggle score" every few seconds
Auto Adaptations
- Text grows larger, contrast deepens, busy layouts simplify
- Zero setup — you never configure anything
- Changes fade in when struggle rises, fade out when comfort returns
Focus & Reading Mode
- A soft spotlight follows the paragraph you're reading
- Ads, sidebars, and popups hide automatically
- Toggle from the popup or say
focus mode
Voice Commands
- Talk to the browser:
scroll down,click Submit,go back,summarize this - 28 languages — all 10 major Indian + English, Chinese, Spanish, Portuguese, Russian, French, Arabic, Indonesian, German, Japanese, Turkish, Vietnamese, Korean, Filipino, Persian, Italian, Thai, Polish
- Speak naturally — common phrases, not rigid syntax
Eye Tracking
- Move the cursor just by looking at the screen
- Powered by your webcam — no extra hardware needed
- 30-second calibration and you're in control
Dwell Click
- Hover over any button — it clicks itself
- A ring fills up while you dwell — always see the click coming
- Move away to cancel — no accidental clicks
Fatigue-Adaptive UI
- Extra panels hide, spacing opens up, colors soften
- Tracks time of day + session duration
- Eases through 4 gentler levels as the day wears on
AI-Powered Insights
- Dense pages → plain-English summaries
- Jargon → simpler words; long emails → bullet points
- On-device AI first (free); Gemini or Claude only for hard requests
Multi-Language
- Type
namaste→ नमस्ते (press Alt+T to toggle) - Voice recognition auto-switches based on page language — detects Devanagari, Tamil, Telugu, Kannada, Bengali, Gujarati, Gurmukhi, Malayalam, Cyrillic, Hangul, Thai, Perso-Arabic
- 28 languages ≈ 87% of the world population
Global Reach
Voice commands and page-language detection across 28 languages — reaching ~6.96 billion speakers, roughly 87% of the world population.
Accent-bordered rows = the 10 Indian languages with full native-script voice-command registries and Latin → Devanagari / Tamil / Telugu / Kannada transliteration (press Alt+T in any input). Non-Indic rows = voice-recognition locale set automatically from page Unicode range (Cyrillic, Hangul, Thai, Perso-Arabic) or from the Settings dropdown.
Architecture
A privacy-first, on-device accessibility operating layer. Behavioral signals are captured in the DOM, scored locally, and a targeted adaptation lands inside 50 ms — no network round-trip, no user configuration, no disclosure required.
Signal → Adaptation Pipeline
Four stages from raw mouse event to applied DOM mutation. Runs entirely inside the Chrome process — the background service worker handles scoring, content scripts do collection and mutation. Nothing leaves the browser for adaptation decisions.
Collect
Five passive DOM listeners sample scroll velocity, click accuracy, typing rhythm, cursor path, and dwell time without blocking input.
Score
Each signal is normalized to 0–1, batched, and fed into an EMA-smoothed weighted composite that produces a live struggle score.
Decide
A rule-based classifier evaluates the score against the user profile and emits one of 16 adaptation types when a confidence threshold is crossed.
Adapt
The adaptation is routed to SensoryAdapter, CognitiveSimplifier, VoiceCommand, or DomainConnector, which mutates the DOM with a reversible handle.
Layered System
Each layer owns a single concern and communicates through a typed message bus. Swap any layer (UI framework, AI provider, signal source) without rewiring the rest.
chrome.storage.local so it survives popup close.PROFILE_UPDATED to every open tab.Monorepo Packages
Three typed packages wired through pnpm workspaces. Every cross-package import goes through an exports map so refactors land atomically.
@accessbridge/core
Types, ProfileStore, StruggleDetector, DecisionEngine, i18n (transliteration rules + language ranges). Zero external runtime dependencies.
@accessbridge/extension
Manifest V3 Chrome extension. React popup + side panel, content scripts, 6 domain connectors, SensoryAdapter, TransliterationController, LanguageDetector.
@accessbridge/ai-engine
3-tier AI with cache, cost tracker, and provider router. Summarizer, Simplifier, Classifier, Translator services — all tier-aware.
Deploy Stack
Dockerized nginx + FastAPI on Hostinger KVM2. Self-hosted update server (/api/version), landing page, extension download.
Technology Stack
Picked for longevity and zero-runtime-dependency where possible. Strict TypeScript end-to-end; no any in production code.
Accessibility Challenges We Solve
The lived barriers AccessBridge was built to dissolve — spanning vision, motor, cognitive, linguistic, temporal, and social dimensions of accessibility. Each pairs a real-world problem with the on-device mechanism that addresses it.
Small text, low contrast, color blindness
Roughly 2.2 billion people live with some form of vision impairment. Yet most websites ship a single 14-16px body size, fixed contrast, and a palette that's invisible to 8% of men with red–green color blindness.
AccessBridge does: detects repeated zooming, slower scroll, and re-reading of the same paragraph, then quietly scales text, deepens contrast, and remaps colors for protanopia / deuteranopia / tritanopia — no site cooperation required.
Can't use a mouse or fine-motor control
People with Parkinson's, ALS, severe arthritis, or hand tremor can't reliably click a 16px button or drag a scrollbar. Dedicated assistive hardware (gaze trackers, switch devices) costs $1,500 – $3,500 and needs setup most users never finish.
AccessBridge does: turns any laptop webcam into a gaze cursor (30-second calibration), adds hover-to-click with a visible progress ring, and runs the browser by voice in 28 languages — all with zero extra hardware.
Attention difficulties & cognitive overload
People with ADHD, autism, traumatic brain injury, or anxiety get swamped by modern web pages where 70–80% of the pixels are sidebars, ads, notifications, and recommendations — all competing for attention they can't spare.
AccessBridge does: a soft spotlight follows the paragraph being read while everything else dims; ads and sidebars hide on demand; layout simplifies automatically when struggle signals rise — the page gets quieter, not louder.
Content & tooling isn't in my language
Over 4.8 billion people speak a non-English primary language. Enterprise apps ship English-first, voice features rarely understand Tamil or Bengali, and Latin→Indic keyboards are a Windows-only pain no one wants to configure.
AccessBridge does: voice recognition in 28 languages (10 Indian first-class + English + 17 global including Portuguese, Russian, Indonesian, Korean, Persian, Thai); Unicode-range page detection sets the voice locale automatically; press Alt+T in any input to type in Latin and watch it become Devanagari / Tamil / Telugu / Kannada.
Fatigue & eye strain over long sessions
Mental fatigue, digital eye strain, and migraine thresholds build up across an 8-hour workday. An interface that felt fine at 9 am is hostile at 10 pm — but no website knows or cares what time you started.
AccessBridge does: tracks time-of-day and session length, then progressively simplifies the UI through four calmer levels — hiding optional panels, opening spacing, softening colors, reducing motion — without the user pressing a thing.
Dense jargon on critical portals
Banking, insurance, healthcare, and legal sites assume every user parses "deductible," "prior authorization," "underwriting," "1.5 Lakh." When you can't, the stakes are real — a missed claim, a wrong premium, a refused treatment.
AccessBridge does: on-device AI rewrites jargon into plain language and summarizes long emails into bullets; 6 domain connectors (Banking, Insurance, Telecom, Retail, Healthcare, Manufacturing) decode terms inline as you browse.
The stigma of asking for help
Many users with disabilities — especially invisible ones like dyslexia, low vision, or cognitive differences — hide their needs at work or school rather than formally disclose. The ask itself carries social and professional cost.
AccessBridge does: zero disclosure. The user never tells the website, the employer, or a coworker. Every adaptation happens silently on the user's own browser; the website never knows an accommodation fired.
Assistive tools cost more than the device
Commercial screen readers run $900 – $1,200. Eye trackers start at $150 and climb past $25,000 for medical-grade. Enterprise assistive suites license at $400/user/year. Most people who need them can't afford them.
AccessBridge does: ships as a free Chrome extension. Runs on any laptop made in the last five years. Uses the webcam you already have, the microphone you already have, and the browser you're already in.
Privacy & Performance Guarantees
Non-negotiables the architecture was designed around from day one.
- 100% on-device for behavioral data — zero network I/O for struggle scoring or adaptation decisions.
- Zero disclosure — the user never tells a website they need help; the website never knows an adaptation fired.
- Reversible by default — every adaptation carries a revert handle. Master toggle off = every mutation undone in one message.
- AI cascade — free local tier first; paid tiers gated by a daily cost budget tracked in-browser.
- Strict TypeScript end-to-end; no
any, no// @ts-ignorein production code. - Adaptation latency < 50 ms from signal threshold crossing to DOM change.
- Content-script syntax-check runs on every build (
node -c dist/…/content/index.js) to catch the chunk-collision trap before it ships. - 187 unit tests cover struggle scoring, decision rules, profile persistence, transliteration, language detection, and the AI cache / cost tracker.
The Full AccessBridge Vision
The browser extension is the wedge. The real platform is the core + AI engine — every surface where accessibility gaps exist is a candidate deliverable.
Same Core, New Surfaces
Everywhere accessibility matters — not just browsers.
- Desktop companion (Tauri) — ambient adaptation for Word, Excel, Teams, Slack, VS Code
- Android accessibility service — system-wide for native apps
- iOS app — Safari extension + in-app SDK
- CLI / Terminal mode — jargon decoder, error-message simplifier
Enterprise Layer
Deploy + monitor accessibility across an organization.
- Admin console — org-wide profile defaults, aggregate struggle heatmaps
- Compliance scanner — WCAG / Section 508 audits with reports
- Profile sync + SSO — SAML/OIDC, profile follows the user
- Audit log + ROI reporting — which apps cause most struggle
Ecosystem (Developer)
Turn the AI engine into a platform, not just a product.
- JS/TS SDK — embed AccessBridge in any web app, no user install
- Public API — summarize + simplify endpoints for integrations
- Figma plugin — audit designs for accessibility before build
- VS Code extension — warn devs writing inaccessible markup
- Domain connector framework — community contribs per industry
Research / Moonshot
Where accessibility intersects with ambient computing.
- Real-time meeting captions + translation — Teams, Zoom, Meet
- AR overlay for smart glasses — menu reading, sign translation
- Voice-first OS layer — full voice navigation for Windows
- On-device ML models — WASM + WebGPU, zero cloud dependency
What is the Observatory?
The Observatory is an organization-wide dashboard that shows how AccessBridge is helping real users across the web. It aggregates anonymized signals — languages people browse in, adaptations that fire most often, struggle trends over time — so teams can see where accessibility actually matters for their users.
Top 5 languages
Top 3 domain connectors
Top 5 adaptations applied
Top 5 features enabled
Struggle events per day
Accessibility score improvement per day
Chrome Extension
The web-facing half of AccessBridge. Runs entirely on-device, no data leaves the browser.
chrome://extensions, toggle Developer Mode, and click "Load unpacked".Desktop Agent New
Extends AccessBridge to native Windows apps via the Windows UI Automation framework. Loopback-only IPC, zero network exposure, encrypted pairing.
Team Deployment Session 24
Roll AccessBridge out to 10 to 1,000 users with a single command. Choose a preset profile, track cohort health in the Pilot Dashboard, and run a structured pilot before org-wide adoption.
Install Command
Installs the desktop agent and loads the Chrome extension for all users. Run as administrator.
pwsh -Command "iwr https://accessbridge.space/team/install.ps1 | iex"
Requires PowerShell 5.1+ and administrator elevation. Downloads and silently installs the agent MSI, then deploys the Chrome extension policy.
curl -fsSL https://accessbridge.space/team/install.sh | bash -s -- --profile=pilot-default
Requires Homebrew and sudo. Installs the agent, loads the mobileconfig Chrome policy, and sets the pilot preset.
curl -fsSL https://accessbridge.space/team/install.sh | bash -s -- --profile=pilot-default
Requires curl and sudo. Drops a Chrome managed-policy JSON file and installs the .deb / .rpm agent package depending on distro.
Preset Profiles
Six ready-to-deploy profiles — each tuned for a specific pilot scenario. Pass --profile=<name> to the install script to activate it.
How to run a pilot
- Step 1 — Choose a preset — Pick the profile that best matches your cohort's needs. You can customise any preset via the admin policy layer before deploying.
- Step 2 — Run the install command — Execute the one-liner above (Windows / macOS / Linux) on each enrolled device, or wrap it in your MDM push tool (Intune, Jamf, Ansible).
- Step 3 — Enrol your cohort — Share the invite link generated by the admin console. Minimum cohort for aggregate metrics is N = 20 (differential-privacy floor).
- Step 4 — Monitor in the Pilot Dashboard — Track enrolment progress, KPI health, and user feedback at observatory/pilot.
- Step 5 — Export & decide — Download the CSV or PDF report at the end of your pilot window. Use it to justify org-wide rollout or refine the preset configuration.
Enterprise Deployment Session 20
Deploy AccessBridge across your organization with Group Policy force-install, signed MSI for the desktop agent, and admin-locked feature profiles. Built for 250,000-employee scale.
ExtensionInstallForcelist. Lock down features per-org with 9 admin-controlled policies (observatory telemetry, AI tier, language, feature enable/disable, minimum agent version, org-hash).AccessBridge on GitHub
Source, issues, and release history for AccessBridge — the ambient accessibility operating layer. Chrome extension, core engine, AI engine, and the Observatory dashboard all live in a single TypeScript monorepo.
Four packages, one TypeScript workspace
A pnpm workspace with strict typing across every package — extension, runtime, AI engine, and shared core all live and ship from the same repo.