Phishing continues to be a critical threat to email security, exploiting human vulnerabilities and bypassing traditional detection systems. This paper presents an AI-powered solution leveraging machine learning and NLP to analyze email content and metadata for phishing indicators. We introduce Cyri, a local-processing conversational assistant that detects semantic cues such as urgency, threats, and emotional manipulation. Cyri’s integration with email clients ensures real-time, privacy-preserving analysis. Experiments demonstrate improved detection accuracy and user comprehension, offering a practical, scalable advancement in phishing defense.