Describe the task, hand over a number, and watch the conversation unfold live — with a full transcript the moment it ends.
Sign in to run a call.
Pick a call type, then tell us who to reach and when.
Fill in what the assistant needs to get this done.
What does a successful call look like? What should be true once the assistant hangs up?
Shape how your assistant comes across on the call.
Tell the assistant what to do when things don't go to plan.
Review everything below. You can preview the assistant in your browser before placing the call.
When you're ready, place the call and watch it unfold live.
Credits never expire. One credit covers one completed call of any length.
| User | Credits | Verified | Joined | Actions |
|---|---|---|---|---|
| Loading… | ||||
The "types of call" users can pick in the wizard. Edit each type's wording and the ElevenLabs prompt variation — the objective the agent is given and the default desired outcome. Add, remove, reorder, or hide types. Changes apply to new calls immediately.
The personas users can pick in the wizard's personality step. Each maps to a talking style, traits, an assertiveness level and extra prompt notes. Add, edit, reorder, or hide them.
The agent must say this, word-for-word, as the very first thing on every call. Use {{name}} for the person the call is placed on behalf of.
Pre-fills new calls and is used when a call doesn't specify its own.
Calls may only START within this window, in the recipient's local time (24-hour clock). The TCPA default is 8 to 21 (8 AM–9 PM). Enforced for both immediate and scheduled calls.
Installs the ask_customer tool on the ElevenLabs agent so it can ask you a question mid-call (you answer from the live call screen). Requires BASE_URL and ASK_CUSTOMER_SECRET set in the environment. Run once.
Pins the agent's TTS stability and language directly on the ElevenLabs agent so its accent stops drifting mid-call. Higher stability (e.g. 0.6) holds the accent steady. Leave language blank to keep the agent's current setting. Note: some newer "expressive" (v3) voices ignore stability.