What's New inNerivo AI

Every surface — the chat header, Settings, the provider pages and the server — now reads from a single model registry. Four separate lists existed before, so the site could advertise a model the chat could not actually select. Entries that no longer exist at the provider have been removed.

Every model was verified against the live provider catalogue, and the new Neri Auto option routes each request to whichever free model is available. A model is only offered when it can genuinely serve a request, so availability is never assumed or invented.

Failures are now classified as rate limit, capacity, timeout, model unavailable, context limit, authentication or network. A capacity problem is reported as a capacity problem, and raw technical status codes no longer appear in the interface — they are recorded in the logs instead.

Temporary failures are retried with exponential backoff and jitter, and a retry instruction from the provider itself is honoured. Failures that can never succeed — a rejected credential, an unknown model, an oversized request — are not retried, so one error cannot become a storm of requests.

When a model is temporarily unavailable, another model can serve the request — but only one that supports everything the request needs. An image request can never be handed to a text-only model. The substitution is always disclosed, and the saved preference is never changed behind your back.

A response in progress can now be cancelled. Cancelling aborts the request instead of leaving it running unseen, and a slow reply can no longer overwrite a newer one.

Typing in the composer previously re-rendered and re-highlighted every AI response in the conversation. Replies are now isolated from the composer, so long answers with large code blocks no longer make typing stutter.

Syntax highlighting loads only the languages in use instead of bundling every supported grammar, through one shared configuration. This removed several hundred kilobytes from the chat page.

An HTTP success status was being treated as proof that a response had been generated. A request can report success while the model itself fails, which produced a misleading busy message carrying a raw status code. Responses are now inspected properly and the real cause is reported.

Several models in the list no longer existed at the provider, including the default model offered to new users — which is why requests could fail immediately. Every entry is now checked against the live catalogue and the dead ones are gone.

Deep Research performs genuine live searches and reads the pages it finds before writing, so answers are grounded in current sources rather than the model's training data.

Every source is resolved to the real publisher URL rather than a search redirect, so each claim can be traced back to the page it came from and checked.

Reasoning and searching are handled by separate, purpose-chosen engines. The search provider is never silently swapped, so a research run is always served by the provider it selected — or it reports the failure instead.

When research stops, Nerivo now says which of three things happened: the research model's free allowance is exhausted, the search provider's allowance is exhausted, or the service could not be reached. A quota limit is confirmed as a limit, not as an absence of information.

The reasoning half of Deep Research can be pointed at an alternative AI engine through configuration, while the search and citation pipeline stays the same.

Research runs on built-in free allowances with no paid search API. Model selection accounts for which engine can actually search for free, so switching engines cannot silently introduce a bill.

A search outage or an exhausted allowance is no longer described as "nothing matched", and an outage is never reported as information that the web does not contain. Outages now say they are outages.

A live activity panel reports each stage as it runs — interpreting the request, planning, searching, reading sources, comparing and verifying — with counters for sources found, inspected, verified and matched. A stage is only marked complete once the work behind it has finished, and every counter is read from real research state.

Each candidate is compared against the requirements supplied. Listings meeting every requirement are marked as exact matches; those with something still unverified are marked partial and the missing item is named; anything matching an exclusion is set aside but still reported, so nothing is silently dropped.

Every finding carries its original publisher link, the type of source, when it was checked and what was extracted from it. Any source can be opened to continue the work independently, and volatile details such as price and stock are labelled as readings that can change.

Page content is handled as data and never as instructions. Text that attempts to issue commands to the assistant is neutralised before it reaches the model, so a web page cannot redirect Nerivo away from the request.

Hold a natural spoken conversation with the AI. Speech recognition runs continuously for the whole call, so every question is captured and answered aloud without touching the keyboard.

Smart voice activity detection distinguishes speech from silence and pauses, finalizing what you said the moment you stop talking for faster, more natural turn-taking.

Cut the AI off mid-answer at any time. Speech synthesis stops instantly and the session switches back to listening, just like a real phone call.

The entire voice session maintains context — the AI remembers what was discussed earlier in the call and answers follow-up questions coherently without repeating itself.

Chunked speech synthesis with keep-alive management reduces perceived latency and prevents long answers from being cut off, delivering smooth, complete spoken responses.

A shared design contract of glass panels, backdrop blur, mint (#00f5d4) and cyan (#22d3ee) accents, gradient text, and glow effects now powers the entire site.

Floating aurora orbs, animated grid, vignette, and 3D card hover transformations bring depth and motion to every page while keeping content fully readable.

A prominent Live Chat Support call-to-action on the homepage connects visitors directly to real-time human support in one click.

Applied the unified theme to all sub-pages — about, features, models, team, contact, help, status, changelog, terms, cookies, accessibility, documentation, and community — with all content preserved.

Custom mint scrollbar, button shine effects, and smooth hover transitions give the interface a polished, cohesive feel on desktop and mobile.

Visitors can start a live chat session that appears instantly in the admin panel, with bidirectional message sync powered by Firestore real-time listeners.

The admin panel is now protected by a login system with a Main Owner account, role-based presets, and custom granular permissions controlling access to every section and action.

The owner can invite teammates by email; invitees accept with a secure email sign-in link and are provisioned with their assigned role automatically.

The owner can add, edit, deactivate, and remove admin users, assign roles or custom permissions, and revoke pending invites at any time.

Session presence and reconnection handling keep both sides connected, with unread indicators and automatic session recovery after network interruptions.

Execute Python, Ruby, C#, SQL, Bash, JavaScript, TypeScript, PHP, Perl, Lua, Swift, Go, Rust, Scala, Elixir, Erlang, Julia, Haskell, D, C, and C++ code instantly with real output.

Run SQL scripts in the sandbox with automatic MySQL-to-SQLite translation — CREATE DATABASE, AUTO_INCREMENT, DATE_ADD, DATEDIFF, and more are handled automatically.

Shell scripts now execute in a real sandbox instead of showing a local-only hint, with smart sanitization for safe inline commands.

Common aliases (js, ts, c#, cs, rb, py, sh…) are normalized automatically, so the right compiler is always selected even when the language name varies.

SQL queries that previously ran but showed no output now render results correctly, with translated syntax errors reported clearly.

Fixed image analysis requests not being properly processed when users attach images and request analysis. Enhanced fallback error handling for Google Vision API failures.

Removed unnecessary rate limit warning popup that was interrupting user experience during normal chat interactions.

Improved error messages for image analysis failures with clear guidance on potential causes and solutions for users.

Advanced profile management combining Firebase Auth with database storage, featuring displayUser object that syncs profile changes across all components in real-time.

Complete Firebase Admin SDK setup with proper authentication token verification, enhanced error handling, and development fallback API for seamless profile operations.

Immediate navbar profile updates across all pages when users modify their profile information, using enhanced AuthProvider with refreshProfile() functionality.

Simplified navbar design with single "Sign In" button for unauthenticated users, removing visual clutter while maintaining consistent authentication entry point.

Unified profile display system ensuring consistent name and avatar presentation across home, profile, and all other pages using displayUser data integration.

Resolved "Failed to update profile" errors with comprehensive error handling, specific error messages for different failure scenarios, and automatic API fallback mechanisms.

Advanced AI-powered code analysis with real-time bug detection, performance optimization suggestions, and automated code quality improvements with support for 50+ programming languages.

Multi-user collaborative coding environment with live cursor tracking, real-time code synchronization, integrated voice/video chat, and shared development workspaces.

Security work covering server-side handling of provider credentials, validation of outbound research requests, and access rules that keep each account’s own data its own.

Real-time application performance monitoring with custom dashboards, automated alerts, error tracking, and comprehensive user behavior analytics with ML-powered insights.

Intelligent documentation generation with automatic API documentation, code commenting, technical writing assistance, and multi-language documentation support.

Optimized mobile interface with gesture controls, offline-first architecture, cross-platform synchronization, and native mobile app features for iOS and Android.

Complete domain verification support for Google Search Console, Bing Webmaster Tools, Facebook Business, and Apple Pay with HTML file, meta tag, and DNS record methods.

Full analytics tracking implementation with Google Analytics 4 and Tag Manager, including environment variable configuration and conditional script loading.

Custom Nerivo AI favicon implementation across all platforms with multi-size support, Apple Touch Icons, and complete manifest.json PWA configuration.

Comprehensive accessibility page with real-time controls (high contrast, font sizing, dark mode), keyboard navigation guide, and screen reader support.

Cross-platform setup scripts (Windows & Linux/Mac) with comprehensive documentation for domain verification and analytics configuration.

Optimized robots.txt, comprehensive sitemap.xml, and structured data implementation for improved search engine visibility and ranking.

Completely restored Firebase authentication functionality with proper module resolution, error handling, and user sign-in/sign-up capabilities.

Added next.config.js with Firebase transpilation support, webpack configuration, and proper module resolution for enhanced compatibility.

Resolved SWC helpers dependencies, improved error handling, and enhanced development server reliability with graceful degradation.

Maintained ChatGPT-like chat interface design for optimal readability and user experience across all chat interactions.

Standardized contact email to help.mahinulislam@gmail.com across all application components and documentation.

Ensured consistent Nerivo AI logo usage (https://i.ibb.co.com/211nDHz7/logo-512.png) across all pages and components.

Comprehensive analytics system tracking all user interactions, performance metrics, and generating actionable insights with data export capabilities.

Extensible plugin system with SDK for developers, plugin lifecycle management, and marketplace integration.

Domain-specific AI assistants for Coding, Writing, Analysis, Design, Research, Strategy, Teaching, and Innovation with tailored expertise.

Advanced learning system that remembers user preferences, conversation context, and personalizes responses for improved interactions.

25+ templates across 6 content types with multiple writing styles, target audiences, and SEO optimization features.

Mobile-first design with offline support, push notifications, and touch-optimized interface for seamless mobile experience.

Native integration with GitHub, VS Code, Docker, and major cloud platforms (AWS, Azure, GCP) for streamlined development.

Advanced model switching between OpenAI, Anthropic, Hugging Face, and Local models with performance optimization.

Universal API support for 12+ services across productivity, communication, data, and cloud services.

Advanced touch controls, speech-to-text optimization, and mobile code editing capabilities.

Comprehensive feedback system with user corrections, continuous learning, and model improvement through interactions.

Color palette generation with 6 harmony types, UI component mockups, and design pattern suggestions with export capabilities.

Advanced voice commands, continuous listening mode, and natural conversation flow with speech recognition.

Like, dislike, and pin important messages with smart conversation management and bookmarking features.

Real-time code execution for Python, JavaScript, and more with interactive debugging and instant output.

Comprehensive customization including temperature control, model selection, theme switching, and voice settings.

Upload and analyze multiple file types including images, documents, and data files with intelligent content extraction.

Military-grade encryption, zero data retention policy, and complete privacy protection for sensitive conversations.

Seamless communication in 100+ languages with real-time translation and cultural context awareness.

Revolutionary multimodal AI that can analyze images, extract text, and provide detailed visual insights.

Significant performance improvements across all AI models with optimized infrastructure.

Enhanced Neri Coder Pro with support for 40+ programming languages and better debugging capabilities.

Improved file upload stability and better error handling for large files.

Your Chat History is now encrypted and secure.

Optimized memory usage for better performance during long conversations.

Pin important messages to keep them accessible throughout your conversation.

Improved conversation search with better filtering and relevance scoring.

Export your conversations to PDF or text format for easy sharing and archival.

Better mobile responsiveness and touch interactions for smartphone users.

Choose from Neri 1,Neri C-1 Coder Pro models based on your needs.

Share conversations and collaborate with team members in real-time.

AI-powered conversation starters and follow-up question suggestions.

Connect your GitHub repositories for code analysis and documentation generation.

Speak your questions directly with built-in speech-to-text functionality.

Pre-built templates for common use cases like code review and creative writing.

End-to-end encryption and zero data retention policy for maximum privacy.

Reduced initial load time by 60% through code splitting and optimization.

Enhanced accessibility with proper ARIA labels and keyboard navigation.

Fixed various rendering and interaction issues on Google browsers.