Testing
=======

Voice Agents can be tested before publishing using the **Simulations** tab.

.. image:: ../images/voice-simulation.png
   :alt: Voice Agent Simulations tab
   :align: center
   :width: 100%

| Two testing modes are available: a text-based **Chat** simulation and a browser-based **Voice Call** simulation.

Opening Simulations
-------------------

Click the **Simulations** tab in the agent header to open the simulations page. You can also trigger a simulation directly from the **Test** button (play icon) in the header toolbar.

The simulation interface loads the agent's current configuration and fetches any call parameters defined in the agent's flow (stage variables or output schema fields). These parameters are rendered as input fields that you fill in before starting a session — they represent the caller context that would normally come from a CSV upload in a scheduled campaign or from the inbound call trigger.

Running a Chat Simulation
--------------------------

Fill in any parameter fields shown (e.g. caller name, account number), then click **Chat**. A chat-style interface opens and sends your typed messages through the agent's full processing pipeline: LLM reasoning, prompt construction, and response generation. STT and TTS are bypassed — input and output are text only.

Use Chat simulation to:

* Verify the agent's LLM configuration and prompt logic
* Test conditional branching in the call flow
* Check that the agent handles edge cases and unknown inputs correctly
* Iterate quickly without needing a microphone or phone

Running a Voice Call Simulation
--------------------------------

Fill in any parameter fields, then click **Voice Call**. The platform initiates a browser-based call using your microphone and speakers. The full STT → LLM → TTS pipeline runs, so you hear the synthesised voice response and the agent hears your spoken input.

Use Voice Call simulation to:

* Verify the STT provider configuration (transcription accuracy, language detection)
* Evaluate the TTS voice quality and settings (stability, speed, model)
* Test end-to-end latency
* Identify turn-taking issues or VAD sensitivity problems

Placing a Real Call
-------------------

The **Place Call** button (phone icon) in the header opens a dialog to trigger an actual outbound call to a phone number of your choice. This uses the agent's configured Twilio number and requires the Twilio credentials to be valid and the agent to be in a deployable state.

Use Place Call to:

* Confirm Twilio connectivity end-to-end
* Test audio quality on a real PSTN connection
* Validate inbound webhook configuration

Post-Simulation Review
----------------------

After a simulation session, the conversation appears in the **Conversations** tab with a full transcript. If the agent was configured for call recording, the audio recording is also available there.

For step-level LLM diagnostics, use the **LLM Traces** tab to inspect the exact prompts sent to the model and the responses received.