Skip to main content
Outbound simulation tests agents that place calls. Tuner can’t dial into an outbound agent, so the flow is flipped: Tuner tells your system who to call, and your agent makes the call to a simulation agent, the AI caller that plays the customer in each case.

​
A SIP URI instead of a phone number

Your outbound agent normally dials phone numbers. Tuner’s simulation agents don’t have phone numbers, they have SIP URIs, the internet equivalent of one: instead of +15551234567, it looks like sip:sim-abc123@sip.usetuner.ai. It works the same way, it’s an address your telephony platform can dial, and most providers (Twilio, Telnyx, LiveKit, etc.) accept a SIP URI anywhere they accept a phone number. So for a simulated call, the SIP URI simply takes the place of the phone number: it goes in the same “who to call” field your API already has, and your platform dials it like any other call.

​
The flow

  1. You give Tuner a way to start a call on your side. Every outbound agent has some API that kicks off a call, on your own server, or on a platform like Vapi or Retell. You provide Tuner with that API call once, and Tuner uses it to start every simulated call.
  2. Tuner builds the scenarios. When you run a simulation, Tuner generates simulation agents, each with a persona, an intent, and something to stress-test, exactly like inbound simulation.
  3. Tuner triggers a call for each scenario. Tuner calls your API and tells it who to call: a SIP URI where the simulation agent is waiting, plus scenario context your agent can use.
  4. Your agent dials, and the conversation happens. From your agent’s perspective this is a normal outbound call, the simulation agent picks up, and the scenario plays out.
  5. The call is linked back to the simulation. The SIP URI that was dialed is the key: when the call reaches Tuner, that URI matches it to the simulation that triggered it, instead of being treated as production traffic.
  6. Your production Evals run. The call is scored against the same Evals, Intents, and Call Outcomes as your live traffic, and the results appear in the Simulations table.

​
Setting it up

How you provide the API call in step 1 depends on where your agent runs:
  • Your own server or stack (LiveKit, Pipecat, or a custom integration), follow the Individual platforms guide.
  • Managed platforms (Vapi, Retell, and others), step-by-step guides are coming soon.

Individual platforms

Set up outbound simulation on your own stack: LiveKit, Pipecat, a custom integration, and more.