Included Health Launches Federated Medical Agent 'Dot' Based on LangGraph and Deep Agents

Included Health unveiled 'Dot', an AI medical guide built on a federated multi-agent architecture powered by LangGraph and Deep Agents. While existing medical navigation systems suffered from rigid UX based on decision trees and delayed emergency recognition, the introduction of LLM-based agents enabled personalized context handling and natural conversations.

Federated Architecture and the Role of Deep Agents

Dot consists of a 'Dot supergraph' acting as the main router and sub-workflows handling domain-specific journeys such as emergency room admissions, appointments, and mental health. Each product team owns and develops parts of the graph in a distributed manner, while Deep Agents' global platform prompts and filesystem-based context management ensure consistency in tone and conversation history across agents. This eliminates the need for users to repeatedly explain information, and shared features like insurance coverage are isolated into platform sub-agents accessible to all agents.

Skills-Based Dynamic Loading and Human Handoff

Clinical capabilities and service definitions are managed via a Skills registry, utilizing Progressive Disclosure so that models receive only brief descriptions initially and load full files when necessary. Third-party employer benefits (20–30 per plan) are also being encoded as skills. Additionally, LangGraph's durable execution feature is used to pause graphs in uncertain situations and route to human member care advocates, implementing a Human-in-the-loop design. Humans serve to 'unblock' agents, and upon returning to the same thread days later, the full context is resumed.

Performance and Clinical Results

The safety of the migration was verified using observability via LangSmith and multi-turn user simulation evals. Following the August client launch, chat engagement rose by 75%, and clinician chat grading showed care recommendation agreement rates significantly exceeding the 95% target. Regular clinical audits confirmed the ability to provide proactive support to vulnerable members by identifying over 99% of high-risk situations.