GPT-Live Voice Smoke Test
Verify GPT-Live Voice over its dedicated native Live endpoint. Do not use azd ai agent invoke, the ordinary Prompt and Hosted Voice helper, /protocols/voice, or a microphone/playback/UI client.
Basic Technical Gate
The smoke is intentionally simple:
Read the deployed GPT-Live Voice agent and latest version.
Connect to:
wss://<project-host>/api/projects/<project>/agents/<agent>/endpoint/protocols/voice/live/sessions?api-version=v1Wait for
session.started.Request a normal close and wait for
session.closed.
Report deployed/active status and the smoke result, then provide the Foundry Portal or Agent Playground URL for real microphone, full-duplex, interruption, voice-quality, and latency validation.
Run the Helper
The helper requires Python 3.11 or later and the verified OpenAI Live SDK:
python -m pip install "openai[realtime]==3.19.2"../../scripts/smoke-test-gpt-live.sh \
--endpoint "$AZURE_AI_PROJECT_ENDPOINT" \
--agent-name "<gpt-live-agent-name>" \
--timeout 180..\..\scripts\smoke-test-gpt-live.ps1 `
-Endpoint $env:AZURE_AI_PROJECT_ENDPOINT `
-AgentName "<gpt-live-agent-name>" `
-Timeout 180The wrappers run smoke_test_gpt_live.py. Use smoke-test-gpt-live.sh or smoke-test-gpt-live.ps1. Use --dry-run / -DryRun to preview without authentication.
Exit codes: 0 pass/help/dry-run, 1 runtime or validation failure, 2 usage error.