Silva, FranciscoSantos, PedroDias, João2026-06-022026-06-022025978-989-758-743-6http://hdl.handle.net/10400.1/29078This paper introduces MentalRAG, a multi-agent system built upon an agentic framework designed to support mental health professionals through the automation of patient data collection and analysis. The system effectively gathers and processes high-sensitivity mental health data from users. It employs locally run opensource models for most tasks, while leveraging advanced state-of-the-art models for more complex analyses, ensuring the maintenance of data anonymity. The system’s models have showed improvements in delivering empathetic and contextually adaptive responses, particularly in sensitive contexts such as emotional distress and crisis management. Notably, an integrated agent for detecting levels of suicidal ideation allows the system to assess and respond sensitively to diverse levels of risk, promptly alerting mental health professionals as needed. This innovation represents a stride towards creating a more reliable, efficient, and ethically responsible mental health support tool, capable of addressing both patient and doctor needs effectively while minimizing associated risks.engArtificial IntelligenceLarge Language ModelsMental HealthRetrieval Augmented GenerationAgentic Workflowconference object10.5220/0013267400003938