How the Voice Agent Works

Edited

This article explains what drives a voice agent's behavior on a call and what Alpharun produces afterward.

Instructions drive behavior

A voice agent's behavior comes entirely from the instructions you author on the Instructions tab — its persona, the context it's given about the caller, the conversation flow you lay out, and the guardrails you set. Alpharun assembles these into the agent's working prompt. You write guidance, not a rigid word-for-word script, and the agent handles the conversation naturally within it.

During a call

The agent:

  • Follows your instructions — its role and personality, the caller context, and the conversation flow.

  • Responds naturally to what the caller actually says, rather than reading a fixed script.

  • Uses any enabled tools to take actions or look things up.

  • Draws on the knowledge base, if one is attached, to answer accurately.

  • Stays within the limits you defined in Behavior and guardrails.

After a call

Alpharun produces a customer interaction record for the call: the recording, transcript, a summary in your chosen format, the classified outcome, and any structured data captured through your data extraction fields. These appear under the agent's Customer interactions, so you review agent calls the same way you review any teammate's calls.

Versioning

Every edit saves to a draft, and publishing promotes that draft to live. Draft and live are tracked separately, so you can make and test changes without affecting active callers until you publish.