As automated agent systems handle increasingly complex tasks, splitting work efficiently between sub-agents becomes essential.
Typically, an agent system relies on a single main model for every sub-task, forcing developers to either pay for an expensive top-tier model on simple steps or rely on a cheaper model that may fail on harder ones.
Resolving this balance requires flexible routing options that direct specific tasks to appropriate models.
Nous Research introduced the 'Advanced Delegation plugin' for Hermes Agent to address this issue.
Expanding on the system's core task assignment features, the plugin adds two routing modes to the delegate_task tool: model routes and profile routes. These options allow developers to assign precise environments to sub-tasks while maintaining isolated execution and background tracking.
The first mode, model routes (model_map), allows sub-tasks to select from a list of user-defined model paths. This lets a main agent route high-volume tasks like basic text editing to fast, low-cost models while reserving heavy coding or multi-step reasoning for higher-tier engines.
The second mode, profile routes (profile_routes), executes sub-agents using established Hermes profiles. Each profile carries its own pre-configured models, credentials, and instructions, allowing teams to isolate work across distinct operational environments.
Developers can configure these modes via the command line with hermes delegation config or during active sessions using the /delegation command.
The plugin also provides strict error handling: if a selected route is misconfigured or loses connectivity, the system halts with a clear error rather than silently defaulting to the parent model.
This transparent feedback offers engineering teams better control when running multi-agent tasks across local and cloud environments.
