What is sara-agent-ai?

SARA-AGENT-AI is a project that builds an AI powered decoy, called Sara, designed to talk to scammers instead of immediately shutting them out. Rather than blocking a suspicious message right away, Sara acts like a real, engaged person, keeping the scammer in a realistic conversation so the system can quietly collect useful details such as phone numbers, UPI payment IDs, bank account details, phishing links, and patterns that describe the scam. The system is built from several connected pieces. When a message comes in, a scam detection engine first checks whether it looks like a scam attempt. If it does, the system pulls relevant past conversation memory using a tool called Hindsight, builds a prompt with that context, and routes the request through a runtime router called cascadeflow, which is meant to pick the right underlying AI model and keep responses accurate. Sara's AI then generates a reply, the reply is checked before being sent, and any threat information found in the conversation, such as phone numbers, bank names, account numbers, URLs, email addresses, the scam category, and a risk score, is pulled out and saved to memory. The project also describes a voice version of Sara, where spoken audio is converted to text, processed by the same AI, and converted back to speech, intended for future use with phone call services like Twilio and Exotel. A dashboard is planned to show live sessions, logs, the memory timeline, extracted threat intelligence, and performance metrics. This is aimed at security teams and researchers who want to study scam behavior and gather intelligence on scammers rather than simply blocking them. Several parts described in the README, including WhatsApp and SMS integration, multi-language support, and voice cloning, are listed as future enhancements rather than features that exist yet. The README does not state a license.