Skip to content

Agents

An agent is the runnable AI experience you configure for Users. In the dashboard, each agent is edited from a detail page with seven visible tabs: Basics, Knowledge, Actions, Controls, Test, Distribute, and Privacy.

Basics covers the agent’s identity and core conversation behavior: name, description, prompt, opening message, language, and runtime fundamentals. Use this tab to make the agent’s purpose specific and easy to evaluate.

Knowledge links workspace documents to the agent. A document in the workspace knowledge library is not automatically available to every agent; attach only the sources the agent should use.

Actions connects the agent to provider-backed integrations, custom actions, built-in in-call actions, and post-session work. Use it when the agent needs to call a tool during a conversation or run a structured follow-up task after a session ends.

Controls manages availability, listening behavior, and per-user personalization inputs. This includes quiet hours, interruption and turn-taking behavior, and profile fields the agent can use in instructions.

Test validates the agent before distribution. Use quick previews for immediate checks and the top-level Tests area for repeatable test suites, example sets, judges, and run history.

Distribute contains the current deployment channels: Web widget, API, Phone, WhatsApp, and Webhooks. Configure distribution from the agent-scoped flow so channel setup stays tied to the correct agent.

Privacy captures agent-level privacy controls. Use it with workspace privacy, compliance, and audit settings for a complete operating posture.