This module of the Hermes Agent Masterclass explains how profiles enable running multiple fully isolated Hermes agents on one machine, each with independent configurations and gateways, while introducing the Conbon board for collaborative, multi-profile task management. It demonstrates a practical use case of a three-profile team working together on a project and highlights the importance of security in managing complex agent workflows.
In this ninth module of the Hermes Agent Masterclass, the focus is on profiles, multi-instance setups, and the Conbon board, which together enable running multiple independent Hermes agents on a single machine. Unlike sub-agents discussed previously, which are ephemeral and share a parent agent’s state temporarily, profiles are fully isolated agents with their own configurations, memories, skills, gateways, and even phone numbers. This isolation prevents state corruption and allows each profile to operate as a standalone agent, making them essential for complex, multi-agent workflows.
Creating and managing profiles can be done through several methods including the command line interface (CLI), the Hermes dashboard web UI, and the desktop app. Profiles can be created from scratch, cloned from existing profiles, or customized with specific models and skills. Switching between profiles is straightforward, either via CLI commands or through the graphical interfaces, and profiles can also be deleted when no longer needed. This flexibility allows users to tailor agents for different tasks or roles efficiently.
Running multiple gateways simultaneously is a key feature enabled by profiles, allowing each profile to connect to different messaging platforms or bots independently. However, it is important to note that two profiles cannot share the same bot token simultaneously, as this causes conflicts and errors. While profiles isolate agent state and configuration, they do not sandbox the system environment, meaning they do not inherently restrict file system access or provide security boundaries. For true sandboxing, additional measures like Docker or SSH-based backends are recommended.
A powerful collaboration tool introduced in this module is the Conbon board, which acts as a durable, multi-profile task routing system. It decomposes a high-level goal into smaller tasks, assigns them to the most suitable profiles based on their descriptions and skills, and manages task progress through various states such as triage, ready, in progress, blocked, and done. The Conbon board supports human intervention for unblocking tasks and persists across restarts, making it ideal for complex workflows that require coordination among multiple agents over time.
The video concludes with a demonstration of a three-profile team—Researcher, Coder, and Writer—working together on a machine learning sports prototype project using the Conbon board. Each profile specializes in different aspects of the project, showcasing how tasks are automatically decomposed, routed, and executed collaboratively. The module wraps up by highlighting the upcoming final module on security, emphasizing the importance of safeguarding powerful agent workflows to prevent potential misuse or damage.