What makes up an agent
- A persona: the name the agent uses for itself on a call (for example, “Hi, this is Grace”) and the voice it speaks with.
- A blueprint: the questionnaire, qualification criteria, scripts, and caller-experience rules the agent follows. See Blueprints.
- Per-agent overrides: an agent can diverge from its blueprint on specific fields, including its system prompt, first message, name, voice, model, which tools it can call, which knowledge-base documents it can draw on, and its handoff routing for transfers. Everything else is inherited from the shared blueprint.
- Phone numbers or campaigns: the inbound numbers routed to the agent, or the outbound campaigns that dial on its behalf. See Phone numbers.
Creating an agent
1
Open Agents or Build
From Agents, select Create agent, or go to Blueprints directly.
2
Choose a blueprint
Pick an existing blueprint to base the agent on, or select Start simple for a prompt-first agent when a full gate flow isn’t needed.
3
Name and configure
Give the agent a name and voice, and adjust any fields you want to diverge from the blueprint.
4
Connect it to a phone number or campaign
Route an inbound number to the agent, or attach it to an outbound campaign.
Editing a live agent
Each agent has its own editor, organized into tabs: Agent (name, voice, model, system prompt, first message), Tools, Knowledge base, Scenarios, Criteria, Dynamic assembly, Guardrails, and Handoff.Edits made in the agent editor apply to the live agent immediately, for calls that start after you save. There’s no draft state at this level: a change either saves and goes live, or it’s rejected. See how publishing works for what runs before a change is allowed through.
Where to go next
- Blueprints: the flow an agent follows, and how changes reach a live agent.
- Voices: choosing or cloning a voice.
- Knowledge base: giving an agent firm-specific answers.
- Tools: what an agent can do mid-call.
- Testing: checking an agent before it takes real calls.