Start the wizard
From the dashboard’s home screen, click New agent. You’ll be offered two paths:- Start from scratch — build everything yourself, step by step.
- Start from a template — pick a ready-made example and adjust it.

Agent: name, greeting, and instructions
Give your agent a name (for your own reference — callers won’t hear it) and write its opening line: the first thing it says when someone connects, something like “Hello, thanks for calling. How can I help you today?” This step also has a toggle for whether the agent should ignore anything the caller says during the greeting, and the system prompt — the instructions that shape how your agent behaves. If you’re not sure where to start, browse the prompt library for ready-made snippets and edit one to fit.
Engine: languages and AI services
Pick the language(s) your agent should speak, then choose which connected provider handles the transcriber (speech-to-text), the language model, and the voice (text-to-speech).Only vendors you connected on the Integrations page show up here. If a dropdown looks empty, go back and connect a vendor for that role first.

Delivery: how callers reach it
Choose how this agent will be reached: a telephony provider like Plivo, or WebSocket — a browser test call, no phone number needed. WebSocket is the default and is good for trying the agent out before spending anything on real calls.
Call: timing and hold behavior
Set call timing and limits, hold phrases the agent uses if it needs a moment, and whether it periodically checks that the caller is still there.
Review and test
Read back everything you’ve set up. When it looks right, click Finish — this creates the agent and opens a live test call right in your browser.