Scale from One Agent to Profiles, Squads, and Model Routing

One Hermes installation can support multiple AI employees. Profiles create separate roles. Squads organize profiles, humans, and other agent runtimes into a team. Routing instructions determine which member should handle each type of work.

The objective is not to maximize the number of agents. It is to separate responsibilities and use the most appropriate model or tool for each job while preserving one accountable leader.

Each Hermes profile can be registered in Multica as a separate runtime. A default profile might act as chief of staff. A personal profile might manage errands and health. A client profile might have access only to a specific workspace and set of applications.

A squad groups these members. One member is designated as the leader. The leader can assign work to human participants, Hermes profiles, Codex, Claude, or other connected agents. Instructions on each member define when that member should be used.

One routing arrangement uses Hermes as the leader because it holds the broadest context. A design and planning model receives work that benefits from visual planning and high-level reasoning. A coding model receives implementation work. Hermes coordinates the sequence, preserves context, and monitors the project through Multica.

Squad members can also have their own skills and environment variables. A coding agent can hold specialized YouTube or development skills. A client agent can hold only the credentials needed for that client's systems.

Different models and harnesses are built for different jobs. Forcing one agent to perform every task can reduce quality and make the system less flexible. Model routing allows the chief of staff to delegate rather than attempt everything directly.

Separation also improves safety and organization. Personal data does not need to sit in a business-operations profile. A customer-service agent does not need trading credentials. A development agent does not need access to private messages.

A clear leader prevents the opposite problem: several agents working independently without a shared objective or owner.

Create profiles by role, not by novelty. Start with the smallest useful set:

1. Chief of staff: broad context, project ownership, routing, and communication.
2. Personal agent: personal tasks, reminders, and private context.
3. Specialist agent: a distinct function such as development, design, support, or content.

Register each profile in Multica with a separate API token when separation is important. Verify that each appears as a distinct runtime.

Create a squad and choose the chief-of-staff profile as leader. Add the human user and the required specialist agents. Give each member concise instructions describing what it should receive and what it should return.

For example, a design member can be assigned interface planning and visual direction. A coding member can implement the approved plan. Hermes can create the project, hand off the design task, pass the approved design to implementation, and keep the issue updated.

Test the routing with a contained project. Inspect which agent received each task and whether the leader preserved a coherent outcome.

This is an advanced layer and is not necessary for most early use cases. Every additional profile creates more configuration, credentials, memory boundaries, and routing decisions. Weak instructions can cause duplicated work or inappropriate handoffs.

A specialist model may be preferred for a task in one setup without being universally superior. Routing rules should reflect observed results in the user's own workflows and should be revised when model performance changes.

Scale by role: keep one context-rich leader, create profiles with clear boundaries, and route specialized work to the agent or model that has the right skills and access.