This study addresses the challenge of real-time knowledge access from dynamic web data, where conventional search engines return raw links without contextual insight. We present Insightflow, a web-based chat companion that integrates Trafilatura for content extraction, HuggingFace embeddings for semantic search, FAISS for vector indexing, and GROQ’s LLaMA 3.3 70B model via LangChain for intelligent, memory-based interaction. Experiments demonstrate high retrieval accuracy (up to 98% for single-URL queries), low latency (<300 ms for semantic search), and strong user satisfaction (89%). The system enables URL-driven contextual Q&A, open-ended dialogue, and session memory. Our contribution lies in providing a scalable, efficient platform for knowledge augmentation that balances accuracy, speed, and conversational continuity. The proposed framework has applications in academic research, business intelligence, and real-time decision-making.