XInsight: Integrative Stage-Consistent Psychological Counseling Support Agents for Digital Well-Being

A research team led by Fei Wang has introduced XInsight, a novel multi-agent AI framework designed to bring structured, clinically-grounded psychological support to web-based digital well-being platforms. The system directly addresses the limitations of current LLM-driven chatbots, which are often opaque and operate as single-stage tools without alignment to established therapeutic practice. XInsight models counseling as a coherent workflow based on the classical Exploration-Insight-Action paradigm, orchestrating a team of specialized AI agents under a unified Reason-Intervene-Reflect cycle.

Key components include an Exploration Agent that organizes user background into a structured Case Conceptualization Form, a Routing Agent that performs Adaptive Therapeutic Routing (ATR) between methodologies like Solution-Focused Brief Therapy (SFBT), Cognitive Behavioral Therapy (CBT), and Mindfulness-Based Cognitive Therapy (MBCT), and a unified Therapeutic Agent that executes school-consistent interventions. A dedicated Recording Agent continuously transforms open-ended dialogues into standardized psychological artifacts—including case formulations and relapse-prevention plans—enhancing accountability and care continuity. To rigorously assess such systems, the team also created XInsight-Bench, which employs a Scale-Guided LLM Evaluation (SGLE) protocol combining therapy-specific clinical scales with general counseling criteria. Experiments indicate XInsight outperforms existing multi-agent counseling systems in paradigm alignment, therapeutic integration, and interaction depth, offering a practical blueprint for deploying accountable, interpretable AI support in mental health applications.