Redesign Video Brain: Step 1=Enter URL, Step 2=Auto-analyze, Step 3=Ask follow-up questions

Add transcript paste feature for Video Brain - users can paste transcript directly

Add web search fallback for Video Brain when YouTube transcript fails due to network issues

Add YouTube transcript extraction, improve Video Brain with follow-up questions, fix all modes

Fix model, improve error handling, add startup diagnostics for all modes

Fix Video Brain URL handling, add API health check, improve error diagnostics

Complete fix: All modes now use Tavily fallback, always show images/sources/links

Major fix: Improved web search with Tavily AI answers, better content extraction, enhanced system prompt

Fix all modes - use WebSearchGraph, fix image format, add trafilatura

Fix LLM model to openai/gpt-oss-20b

Fix backend URL for Hugging Face deployment, improve startup health check

Deploy to Hugging Face Spaces - Remove Render, enable full features

Fix: dos2unix for line endings in start.sh

Fix: LangGraph node name conflicts and port binding

Fix: Correct Groq model openai/gpt-oss-20b

Fix: Groq model, error handling for all modes, LITE_MODE fallbacks

Fix: Force LF line endings for shell scripts

Fix: Separate start.sh for port 10000

Fix: Video Brain auto-load URL and Enter key submit

Fix: Tavily API format and Streamlit image compatibility

Fix: Streamlit on port 10000 for Render

Fix: Use port 10000 for Render (Streamlit as primary)

Fix: Use openai/gpt-oss-20b model

Fix: Add LITE_MODE for Render free tier (512MB RAM)

Fix: Use correct API keys (Groq + Tavily only)

Add Product MVP and Video Brain modes + Render deployment

Fix: LangChain 0.2.x compatibility - update imports and model config

Fix: Compatible langchain versions + better startup script

Fix: Lightweight requirements for Azure - CPU-only PyTorch

Add Docker and Azure deployment configuration

Add or update the Azure App Service build and deployment workflow config

Perplexity AI Clone - Full Production Version with 8 Modes