ONE Voice Lab · Experimental · Simulated

Simulation Lab

Change bounded synthetic conditions, watch failure propagation, and compare evidence without pretending it is production telemetry.

Deterministic replayNo provider spendOne run is not a benchmark

Agent Access

Human-authorized local replay

WebMCP awaiting detection

Exactly four browser tools share this visible Simulation Lab state. Replays are deterministic, memory-only, and use zero provider requests, credentials, microphone access, telephony actions, or provider spend.

list_voice_scenarios
prepare_voice_replay
run_voice_replay
get_voice_replay_evidence

Prepared plan

No agent-prepared plan is active.

Authorization: noneThe agent cannot click or invoke this human-only transition.

Recent in-memory activity

No agent or authorization activity yet.

Latest deterministic evidence

No authorized replay has completed in this tab.

Example request

Inspect ONE Voice Lab and help me test a contact-center voice agent under crosstalk. Prepare a deterministic local replay, wait for my authorization, run it, examine the evidence, and recommend the next experiment. Do not use live providers.

Scenario registry

Stress the system deliberately

11 defined scenarios

Flagship experiment - Implemented

Target Speaker vs. The World

A flagship deterministic replay for target speech, crosstalk, environmental noise, interruption, silence, and recovery.

ObserverPublic replay only

ModeArchitecture templateControlled impairmentRun number

Audio
STT
Conversation
Agent
Tool
TTS
Playback
Outcome

Realtime view

Event timeline

IDLE

No run is active

Opening this page made zero provider requests. Start the deterministic replay explicitly when ready.

Owner-controlled usage

Live provider mode remains gated

Public visitors can inspect definitions and run local replay. They cannot launch live provider matrices or spend Deepgram credits.

Provider requests: 0
Audio submitted: 0 sec
Billing: Usage captured. Billing value not estimated.