voicebot
engine: …
Bots
Calls
identity
Name
System prompt (persona / instructions)
Exotel stream URL (point your applet here)
⧉
test in browser
🎤 Test
idle
native_rs engine URL (this browser connects here)
Runs a real mic call from this browser. Save the bot first. Allow mic access when prompted.
pipeline
Form
cascaded (STT→LLM→TTS)
speech-to-speech (realtime S2S)
audio-in LLM (audio→LLM→TTS)
Engine override
(process default)
pipecat
livekit
native_py
native_rs
Forms 2/3 run on the native engines only (native_py / native_rs) — pipecat and livekit are cascaded-only.
S2S — realtime session (form 2)
Provider
nova_sonic
openai_realtime
Model
(blank = provider default)
Voice
(blank = provider default)
Shadow STT (ship's-log transcript)
on (default)
off
Audio-in LLM (form 3)
Provider
openai_audio
gemini
Model
(blank = provider default)
The model hears the caller's turn natively and replies in text — the reply speaks through the TTS block below. Tools and llm-generated greetings are not supported on this form.
STT — speech to text
Provider
deepgram
sarvam
azure
Model
(tunable)
Language param
Expected languages (comma sep)
LLM
Provider
anthropic
bedrock
Model
(tunable)
Max tokens
Temperature
TTS — text to speech
Provider
sarvam
elevenlabs
Voice / speaker id
TTS model
(e.g. bulbul:v2 / bulbul:v3)
Language
(Sarvam target, e.g. hi-IN)
Sarvam speakers — v2: anushka, abhilash, manisha, vidya, arya, karun, hitesh · v3: shubh, aditya, ritu, priya, rahul, kavya… (speaker must match the model)
VAD — voice activity (Silero)
Confidence
Start secs (attack)
Stop secs (hangover)
Min volume
End-of-turn wait (s)
(lower = snappier, may cut off)
greeting
Mode
fixed text
llm-generated
none (caller first)
Reply language
mirror caller
fixed
Greeting text (used when mode = fixed)
timers
Idle stage 1 — nudge (s)
Idle stage 2 — give up (s)
Max duration soft cap (s)
Hard grace over soft (s)
Save
Delete
Select a call on the left to see its transcript, audio, latency and event timeline.