An assistant is either the attacker your red team calls people with, or the defender that protects a phone line.
Calls your people in a cloned voice and plays a persona, to test whether they can be fooled.
Runs on the DeepBlocker calling pipeline from your campaigns. Used with consent, inside authorised exercises.
Answers a phone line, screens who is calling, and checks for synthetic voices before anyone is put through.
Deepfake screening, security questions and verified callers are set up on the assistant page after creation.
Who the attacker claims to be, and how they open the call.
Shown in your campaign list, never spoken.
Short sentences and a clear goal work best. The pipeline keeps replies phone-length on its own.
Optional. Research the real person this persona impersonates from public sources, so the caller can answer who-are-you questions on any attack type.
Said the moment a person answers. Voicemail and phone menus get nothing.
The voice, including your cloned voices and their engine (ElevenLabs or the self-hosted fallback), is chosen on the assistant page right after you create it.
Campaign calls run on the DeepBlocker calling pipeline. The one-off test call from the assistant page still runs on Vapi.
Documents and web pages the assistant can draw on during a call: company facts, names, procedures.