Prompt, model, voice, tools, knowledge and flows side by side, with the cost per minute and latency in view before a single customer calls.
Cost & latency, liveFunctions the agent can callTalk to it in your browserBrowse ElevenLabs and Deepgram voices with filters for gender and accent, preview them, or add a custom or cloned voice. A fallback voice keeps calls going if a provider has an outage.
Call your HTTP endpoint, transfer to a person, press keypad digits, send an SMS, end the call or use an MCP server. The agent decides when, and says something while it waits.
Split a conversation into steps, each with its own instruction and knowledge, and let conditions decide where the call goes next. Validate the flow before you save it.
Edits stay in a draft until you publish. Each publish is a version you can restore, and the browser test uses your draft prompt so you hear changes before callers do.
Pick the language model, with a fallback model.
Speech recognition per language, with a fallback transcriber.
Web pages, files and text the agent can look up.
Response wait, interruption sensitivity and reminders.
Summary, sentiment, outcome and your own fields.
PII redaction, retention and safety settings.
Sign up in a minute, describe it to Atelier and test it in your browser. Prefer to build it together? Talk to us.
