Changelog
Tracking the evolution of Vyoma and Arka AI.
A detailed record of updates, improvements, and milestones as we build the future of intelligent systems.
DiaCare AI — Vyoma's second product
A friendly AI companion for diabetes education, risk checks, food guidance, and report explanation. Built on Arka Helix 1, free to use, tuned for Indian context. Lives at diacare.vyomagroup.online.
- Helix 1 system prompt tuned for diabetes — 6,500-character prompt with Indian-context specialisations, IDRS risk scoring, conservative refusal patterns for diagnosis and prescription
- Local-first health profile — user health data lives in browser localStorage, never on our servers
- Multilingual — English, Hindi, and Hinglish natively; Tamil/Bengali/Marathi on the roadmap
- Full documentation — 45 articles covering chatting, health profile, diabetes education, food, daily management, medications, safety, and privacy
- Safety guide — comprehensive privacy and emergency-escalation policy
- 20 new blog articles on diabetes, accessible from the Vyoma blog
- 40 research entries on diabetes + AI methods, accessible from the new Research page
Arka Helix 1 — the biggest release yet
A foundational rebuild of Arka. New model identity, a complete Projects system, a Claude-style pricing page, native-feeling mobile chat, and a cache pipeline that keeps every user up to date automatically.
- Arka Helix 1 — renamed the default model from "Arka Standard" to Arka Helix 1, our first named model. Helix 1 is a carefully-tuned wrapper that routes through high-quality open-weights inference, layered with our personality, memory, and safety system prompts
- Adaptive Thinking moved into the model selector dropdown — one tap flips Arka into deep-reasoning mode for harder problems
- Projects — a full project system in the sidebar: create projects, attach chats, browse cards, with a default "How to use Arka" guide seeded for every user
- Claude-style pricing page — new full-screen "Plans that grow with you" overlay with Individual (Pro $7/mo, Max from $21/mo) and Team & Enterprise tabs
- Plus menu overhaul — one + button for Add files, Take a screenshot, Add to project, Web search toggle, and a Use Style submenu (Normal / Learning / Concise / Explanatory / Formal)
- Tips popup — Kimi-style friendly error nudge when the model returns 500/429/timeout, with Got it and Upgrade actions
- Get Arka apps — download icon next to the user profile opening a modal with Desktop, Android, iOS, and Chrome Extension cards (all Coming Soon)
Mobile chat box, properly fixed
Closed out a long-running mobile clipping bug and rebuilt the input card to feel native on every phone.
- Replaced the legacy single-row input with a vertical card: textarea on top, single-row action bar below
- Fixed input clipping caused by stale
100vhoverriding100dvh— now uses100svhso the input always sits inside the visible browser chrome - Forced a strict single-row layout for the action bar on mobile so + / model / mic / send never stack vertically again
- Added safe-area-inset padding for devices with home indicators
- Sidebar trimmed: Export and Clear All removed; Search, Chats, and Projects nav links added in their place
Updates that actually reach users
A bulletproof cache-invalidation pipeline so every deploy lands on every device, without anyone clearing browser data.
- Service worker
CACHE_VERSIONis auto-bumped on every build viabuild.ps1 - HTML shells now use a network-first strategy — users always get the latest UI
- New SW posts
SKIP_WAITINGand the page hard-reloads once oncontrollerchange - SW update probe tightened from 30 min to 5 min, plus an extra check on every
visibilitychange - Build script now emits a ready-to-upload
public_html/folder so deploys are drag-and-drop
Usage tab, in percentages
A cleaner home for daily limits and a brand-consistent take on the entire app's feedback layer.
- New Usage tab in Settings — long progress bars per resource (Images, Deep Reasoning, Files) with percentage labels instead of raw "X / 5" counts
- Removed the top-bar usage badges and the emoji "Today's Usage" mobile dropdown — everything now lives in one tab
- Success toasts rebranded to Arka orange instead of generic green
- Model selector moved into the chat box as a rounded-rectangle chip, no more pill
- Removed the green "live" dot before the model name — visual noise we didn't need
No more stuck splash screens
Mobile users who hit the boot-time splash screen on a slow network would occasionally see it forever. Not anymore.
- 6.5-second failsafe in
index.html— if the Firebase auth boot stalls, the loader cleanly hands off to the login screen - Patched a missing closing brace in
_modal.cssthat was silently swallowing later mobile CSS during minification - Added mobile-only display guards so hidden modals can't widen the viewport
- Production CSS now ships with a versioned URL like JS, so stale Chrome caches can't serve old styles
- Explicit
setPersistence(browserLocalPersistence)on Firebase Auth so F5 / new tabs never sign users out
Constitution and Public Pages
Expanded the Vyoma Group website with new public-facing pages that communicate our mission, values, and product philosophy to users and stakeholders.
- Published Vyoma's Constitution outlining core AI safety principles and ethical guidelines
- Launched the About page with company story, vision, and differentiation
- Introduced this Changelog page for transparent product tracking
- Added Constitution link to the Company navigation dropdown across all pages
- Implemented consistent sub-page design system with shared typography and spacing
- All new pages follow the established Anthropic-inspired minimal aesthetic
Homepage Redesign and Code Architecture
Complete overhaul of the Vyoma Group homepage with a focus on minimal design, professional aesthetics, and clean code separation.
- Redesigned homepage with Anthropic-inspired minimal editorial layout
- Replaced all emoji-based icons with SVG assets from a curated icon library
- Separated monolithic HTML into clean architecture: index.html, style.css, main.js
- Implemented IntersectionObserver-based fade-up scroll animations
- Added responsive breakpoints for mobile (480px) and tablet (900px) viewports
- Introduced CSS custom properties for a maintainable design token system
- Built full navigation system with dropdown menus and mobile hamburger overlay
File Analysis and Usage Tracking
Integrated comprehensive file handling capabilities into Arka AI along with a transparent usage dashboard for managing resource consumption.
- Added PDF text extraction and analysis for uploaded documents
- Implemented support for code files, markdown, and plain text formats
- Introduced 10MB file size limit with clear error messaging
- Built a 2x2 usage dashboard tracking images, deep reasoning, files, and total chats
- Implemented 24-hour rolling usage limits for free and pro tier users
- Added drag-and-drop file upload support with preview chips in chat
- Enabled vision-native analysis for image files using the primary model
Authentication and Stability Fixes
Resolved critical authentication issues and improved overall application stability for deployed environments.
- Fixed Google Sign-In redirect loop on Hostinger-hosted deployment
- Reordered authentication initialization to await getRedirectResult before onAuthStateChanged
- Implemented cache-control headers to prevent browser caching issues during auth flow
- Added .htaccess configuration for proper header management on shared hosting
- Suppressed PHP error output to prevent JSON parsing failures in API responses
Model Standardization and Mobile UI
Streamlined the AI model infrastructure to a single, high-quality model and completed a full mobile responsiveness overhaul.
- Standardized on a single NVIDIA-hosted model for consistent output quality
- Removed multi-model selection UI to simplify the user experience
- Cleaned up backend API to route exclusively through the primary endpoint
- Completed mobile UI overhaul for 320px to 428px screen widths
- Implemented slide-in sidebar with overlay for mobile navigation
- Optimized touch targets and input bar layout for mobile interaction
- Converted settings modal to horizontal tab interface on small screens
Chat History and Settings Panel
Introduced persistent chat history with management capabilities and a comprehensive settings interface for user preferences.
- Implemented localStorage-based conversation persistence with automatic saving
- Added sidebar context menu for starring, renaming, and deleting conversations
- Built settings panel with tabs for capabilities, memory, and tool configuration
- Introduced local-first memory system with import and management features
- Replaced all remaining emoji-based UI elements with proper SVG iconography
- Added keyboard shortcuts for common actions throughout the interface
Multi-Model Integration
Expanded AI capabilities by introducing support for multiple models within a unified conversational interface.
- Added support for multiple AI models within a single unified interface
- Enabled seamless model switching without losing conversation context
- Introduced intelligent model selection based on query complexity
- Improved backend orchestration layer for multi-provider routing
- Added profile avatar system with Google account integration
Introducing Arka AI
The first public release of Arka AI, Vyoma Group's intelligent assistant built for reasoning, learning, and discovery.
- Initial release of Arka AI with core reasoning and conversational capabilities
- Minimal, distraction-free chat interface focused on clarity and usability
- Firebase authentication with Google Sign-In for secure access
- Dark mode support with system preference detection
- Foundation architecture designed for extensibility and future expansion
- Deployed on Hostinger with PHP backend and static frontend
What has improved over time
Deeper chain-of-thought capabilities for complex problem solving
Persistent context awareness across conversations and sessions
Cleaner interface with every update, removing friction at each step
Faster response times and improved reliability under load
Why we maintain a changelog
We believe that building AI responsibly requires more than good intentions. It requires openness about what we are doing, how our systems are evolving, and what decisions we are making along the way. This changelog is one part of that commitment.
Every update we ship reflects choices about what to prioritize, what to defer, and where to invest our limited resources. We want those choices to be visible. When something breaks, we want you to know we fixed it. When something improves, we want you to understand what changed and why.
Trust is not built through marketing. It is built through consistency, honesty, and a willingness to show your work. This page is a record of that work.