Voice & text agent test bench
Hear — and read — the difference between AI stacks before you commit.
Speak or type to an agent, swap the transcription, model and voice provider, and watch latency and cost tick up on every turn. Test in full voice or text-only mode, and run two stacks side by side.
turn 04 · voice-to-voice710 ms
STT
180 ms
Speech to text
LLM
320 ms
First token
TTS
210 ms
First audio
Live voice & text
Talk through your mic or just type — the agent replies as it thinks, sentence by sentence, spoken or written.
Swap providers
Lovable AI, OpenAI, Groq, Deepgram and ElevenLabs for each step — or skip speech entirely in text mode.
Real metrics
Time to first token, time to first audio, voice-to-voice latency and cost per step.
Compare
One message, two differently configured agents answering side by side — in voice or text.