MindfulAgents: Personalizing Mindfulness Meditation via an Expert-Aligned Multi-Agent System

A research team from Carnegie Mellon University and the University of Washington has published a paper on MindfulAgents, a novel system that tackles the persistent problem of user disengagement in mindfulness apps. The system uses a multi-agent architecture powered by large language models to automate the personalization of guided meditation. Instead of offering generic scripts, it generates content based on an expert-established mindfulness framework, prompts users to reflect on their emotional state and skills, and then tailors the meditation experience in real-time for each individual.

In a formative lab study with 13 participants, MindfulAgents demonstrated significant, measurable improvements. It boosted in-session engagement (p=0.011) and self-awareness (p=0.014), while also reducing users' momentary stress (p=0.020). The results were even more promising in a real-world, four-week deployment study with 62 users. This longer trial showed a notable increase in long-term engagement (p=0.002) and participants' overall level of mindfulness (p=0.023). Users reported that the AI-driven system provided more relevant sessions personalized to their specific needs and contexts, which was key to supporting sustained practice.

The research, accepted for presentation at the prestigious CHI 2026 conference, highlights a scalable solution to a major bottleneck in digital mental health. By leveraging LLMs to automate expert-level personalization—a task traditionally requiring costly manual effort from therapists or coaches—MindfulAgents points toward a future where mental wellness tools can be both deeply individualized and widely accessible. This work bridges the gap between evidence-based therapeutic frameworks and the scalable, adaptive power of modern AI.