Hermes Agent

Multi-Agent Orchestration

Hermes Agent is the agent execution engine. It brings together tools, MCP connectors, memory and gateways to the command line, the dashboard and notifications. It acts directly on the system.

Key points: Execution engine · MCP connectors · Persistent memory · Model routing · Gateways and notifications · Skills and sub-agents.

Execution engine

Hermes is the loop that turns a model into an agent. It presents tools to the model, executes tool calls, manages context and approvals, then acts directly on the system. Every step remains traceable.

MCP connectors

Hermes integrates private MCP servers and gives agents access to memory, web search, wiki recall, diagram generation, application control and code indexing, without dependency on a third-party service.

Persistent memory

The agent relies on three-layer memory. It records durable facts and recalls them from one session to the next, which avoids having to re-explain everything in each conversation.

Model routing

Hermes delegates the model choice to LiteLLM and the semantic router. The task goes to a sovereign local model or to an external API depending on sensitivity and the type of work.

Gateways and notifications

The agent is accessible from the command line, from a dashboard and by messaging. It triggers tasks, tracks their progress and notifies results remotely.

Skills and sub-agents

A library of reusable skills encodes proven procedures. Hermes delegates to specialized sub-agents to parallelize research, analysis and writing.