Configurable enterprise AI agents with skills, policies, and MCP tools

No. Although they can converse, they are designed as operating agents: they consult context, follow policies, use authorized tools, generate recommendations, and can trigger workflows with traceability.

A skill is a specific capability an agent can execute, such as classifying a request, extracting obligations, summarizing documents, prioritizing prospects, or preparing a communication.

A policy defines limits and rules: what an agent can do, when it must request approval, which sources it can use, which actions are blocked, and how exceptions are escalated.

MCP tools let an agent interact with specific systems, data, or actions under control. On Cloud they are controlled; on Dedicated and Enterprise they can be enabled or built in greater depth.

The orchestrator evaluates intent, context, permissions, vertical, knowledge source, and policies to select the right agent or workflow. In complex cases it can coordinate several agents.

Yes, if the policy allows it. Some actions can be automatic; others require human approval. The configuration depends on risk, the type of action, and the deployment model.

Executions can log the request, sources used, tool calls, decisions, approvals, user, status, and result. This helps operate with traceability and enterprise control.

Cloud uses predefined agents. Dedicated allows configurable agents with Arvynta support. Enterprise allows deeper customization, an owned agent graph, and advanced governance.