gvox-voice test console

Agent

Mode

STT (cascaded)

LLM (cascaded)

TTS voice (cascaded mode)

Flat execution: whole workflow in one prompt (one LLM call per turn, fastest; step handoffs are prompt-guided) Agent source Single prompt Greeting (spoken first, optional)

Variables ({{template}} values)

first_name

event_name

event_date

event_time

event_venue

Session

Idle. Allow the microphone when asked; the agent speaks first.

Latency · end of your speech → agent audio

—

waiting for the first reply…

session

—

greeting

—

client ≈

—

Transcript

After the call: record from the engine

status

—

Place an outbound call

To (E.164) Provider Agent source Webhook URL (optional; e.g. scripts/webhook_sink.py)

Live status

call

—

Say something on the live call (operator whisper to the caller)

Recent calls

id	dir	status	dur	cost	outcome

Call detail

Select a call.

Workers

id	url	sessions	status	seen

Capacity

engine

—

Webhook deliveries select a call in the Calls tab

seq	event	status	attempts	last

Telephony smoke (fake carrier → real worker endpoint)

Internal token (X-Internal-Token; API must run with ENABLE_TEST_ENDPOINTS=true)

calls

script

Spends real vendor minutes (about 1 per call).

#	call	status	turns	client p50	engine total	ok