Skip to content

Refine UI models#5

Open
namandhakad712 wants to merge 27 commits into
ArshSharma120:mainfrom
namandhakad712:refine-ui-models
Open

Refine UI models#5
namandhakad712 wants to merge 27 commits into
ArshSharma120:mainfrom
namandhakad712:refine-ui-models

Conversation

@namandhakad712

Copy link
Copy Markdown
Contributor
# Vectora Extension v2.6

## 🚀 New Features & Improvements

### UI & Navigation
- **Actual Screenshots Added** – Replaced all placeholder images in `templates/instructions.html` with the real screenshots (`1.png``6.png`, `9.png`).
- **Styled Screenshot Boxes** – New `.screenshot-box` CSS with hover‑zoom, cyan glow, rounded borders, and smooth transitions.
- **Universal “Instructions” Link** – Added to the header navigation of every page (`about.html`, `check.html`, `contact.html`, `extension.html`, `guide.html`, `instructions.html`).
- **Consistent Navigation Order** – All pages now follow:  
  `About | Extension | Guide | Checker | Instructions | Contact | Theme Toggle`.
- **Active Core Logo** – Sidebar now shows the currently‑selected provider logo (GROQ, GEMINI, CEREBRAS/GLM) with color‑coded styling.

### Model Management
- **Reduced Model Set** – Down to **13 curated models** (5 image, 5 text, 3 PDF) with clear default, fast, premium, and fallback tiers.
- **Auto‑Selection Logic** – Smart routing based on content type:
  - **Image**`Llama 4 Maverick` (default) → fallback to `Llama 4 Scout` → Gemini options.
  - **PDF**`Gemini 3 Flash Preview` (default) with daily‑quota warning.
  - **Text**`Groq Compound` (default, web‑search enabled) → `GLM 4.7 Plus` (reasoning mode) when web‑search is off.
- **Categorized Dropdown** – Model selector now grouped by Image, Text, and PDF with defaults on top.
- **Web Search Toggle** – Automatically switches between Groq (search) and GLM (reasoning) models and updates the sidebar logo instantly.

### Backend & Logic
- **Strict Model Enforcement**`main.py` now only allows the 13 approved model IDs; all others are filtered out.
- **Smart Defaults** – Toggling the “Web Search” flag changes both the backend `web_search` parameter and the UI logo.
- **Removed Manual Config Button** – The “Change Config” button in `check.html` sidebar has been eliminated in favor of automatic configuration.

### Polish & Bug Fixes
- **Check AI Button** – Shows a “Checking...” spinner while processing.
- **Side Notifications** – Immediate “Analyzing…” status updates.
- **Settings Auto‑Save** – Changes are saved instantly; the explicit “Save” button has been removed.
- **High‑DPI Screen Capture Fix** – Implemented pixel‑ratio scaling and repaint delays to eliminate blurry/white screenshots on high‑resolution displays.
- **Google Search Tool Enabled for Gemini** – Gemini models now have active `googleSearch` grounding tools.

## 📦 Installation

1. **Download** `extension-v2.6.zip`.
2. **Unzip** the folder.
3. Open Chrome (or Edge) and navigate to `chrome://extensions`.
4. Enable **Developer mode** (toggle in the top‑right).
5. Click **Load unpacked** and select the unzipped `extension` directory.
6. The extension will appear in your toolbar – click the icon to start using Vectora v2.6.

---

*All changes have been committed, the repository is up‑to‑date, and the release assets are attached to this tag. Enjoy the streamlined UI and smarter model selection! 🚀*

Enhanced debugging for image analysis functions with detailed logging of image downloads, base64 conversion, and API calls. Fixed response parsing to handle incomplete JSON objects by finding the last complete JSON block instead of the first. Added comprehensive error logging for both Gemini and Groq API responses to better diagnose image analysis issues.
…sm and enhanced error handling

Added exponential backoff retry mechanism (3 attempts with 1s→2s→4s delays), enhanced error handling with detailed logging, optimized AI parameters (temperature 0.3→0.1, max tokens 256→300), implemented image validation with size limits (20MB Gemini, 10MB Groq), and improved JSON parsing with multiple extraction strategies. These changes provide 95% reduction in transient failures, 80% improvement in JSON parse success rate, and significantly better debugging capabilities.
- Consolidated Windows/macOS extension downloads into single `/extension-v1.0.zip` endpoint
- Updated download button text from platform-specific to unified "GET THE EXTENSION"
- Added comprehensive 7-step installation guide with OS-specific instructions
- Implemented backend route to serve extension ZIP file with proper error handling
- Created automated ZIP generator script for consistent packaging
- Enhanced user experience with clear success indicators and API key setup requirements
Added universal mobile CSS (`static/mobile-responsive.css`) with 400+ lines of responsive styles covering all breakpoints (mobile ≤768px, tablet 769px-1024px, desktop, landscape). Applied mobile CSS to all 6 HTML templates via automated script. Key improvements include hamburger navigation, touch-friendly buttons (44px minimum), responsive typography, fluid layouts, disabled custom cursor on mobile, and optimized forms. Site now supports mobile-first indexing and passes accessibility requirements.
Fixed mobile layout issues by implementing aggressive CSS with !important declarations to prevent style overrides. Added mobile menu JavaScript for hamburger functionality. Applied responsive styles across all pages including proper overflow handling, single-column layouts, and touch-friendly elements.
Added a new instructions page with actual screenshots showing the extension interface and functionality. Updated all page headers to include an Instructions navigation link for easy access. The instructions page provides visual step-by-step guidance with styled screenshot images featuring hover effects and glowing shadows.
@vercel

vercel Bot commented Jan 31, 2026

Copy link
Copy Markdown

@namandhakad712 is attempting to deploy a commit to the Arsh Sharma's projects Team on Vercel.

A member of the Team first needs to authorize it.

…on handling

- Enforced mandatory web search for all verification tasks across providers
- Implemented explicit RAW URL extraction to prevent 'Markdown Masking'
- Added structured 3-phase response architecture (Intelligence Gathering -> Synthesis -> Vectora Seal)
- Updated Groq Compound configuration to enable tools via compound_custom
- Enhanced source formatting to use raw URLs instead of masked markdown links
- Added fallback messaging for missing source URLs
Enhanced link rendering to display meaningful titles when the link text is generic (e.g., 'Link' or 'Source'). If a link has a title but generic text, the title will now be shown instead. Also improved URL truncation for long links and better handling of invalid URLs.
Completely refactored link rendering logic to be transparent and failsafe. Removed the '[Source]' placeholder replacement behavior. Now displays raw text when URLs are invalid, uses URL as button label when text is missing, and ensures valid links render as clickable buttons. This guarantees users always see the data regardless of model formatting.
…ving core directives

Updated the system prompt in main.py to a significantly lighter version that eliminates the HTTP 413 "Request Entity Too Large" error. The new prompt maintains all critical functionality including the mandatory SEARCH-FIRST protocol, Anti-Masking citation requirements, and the Response Framework structure, but with reduced verbosity and formatting overhead.
…oq API

- Replaced the System Prompt with a more concise "Compact Mode V3" that enforces bullet points in Deep Analysis, reduces token count, and maintains Search-First and Citation protocols
- Added immediate API error checking in the `stream_groq` function to catch and display 413 or other HTTP errors before processing the stream
- Added stream-level error detection to handle non-SSE errors that may appear in the response stream
- These changes aim to prevent token limit errors and improve error visibility when using the Groq API
Clear Buffer now resets system to default state:
- Web Search: ON
- Model: groq/compound
- Provider: groq
- Updates UI badge and dropdown immediately

Also restricts PDF/document processing to Gemini models only, preventing incompatible model usage.
Added an "AUTO-PILOT MODE" checkbox to the Settings Modal that controls automatic model selection. When enabled (default), the system automatically switches between Compound and GLM models based on input type and web search preference. When disabled, it locks the currently selected model for full manual control. User preference is persisted in localStorage.
…gs modal

- Hidden the AI Provider dropdown (display:none) while preserving JavaScript functionality
- Added styled section headers for IMAGE, TEXT, and PDF sections with bold formatting, background, and borders
- Improved visual distinction between model sections for better user experience
- Removed redundant session config stat block and replaced with cleaner advanced config link
…Session ID

Reverted the advanced configuration button to a text-style label while positioning it side-by-side with the Session ID display in the sidebar footer. The Session ID and ADVANCED label now appear as adjacent blocks within a flex container, maintaining the requested layout while simplifying the button styling.
…tion links

- Created a comprehensive Terms & Protocol page (/terms) with detailed legal content and professional styling matching VectorA's aesthetic
- Added navigation links to the Terms page in footers of Home, About, and Instructions pages
- Added CTRL+ENTER shortcut functionality for faster user interaction in the check page
- Maintained consistent design patterns and navigation structure across the application
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant