Transcript
Response details
Realtime microphoneReady
Realtime settings
Realtime transcript
Auto detectionFinal
API status
OpenAI-compatible runtime endpoints
Select Refresh to query the API.
System
Inference readiness and GPU runtime
Forced aligner
Checking…
Word and caption-style segment timestamps use an additional multilingual Qwen3-ForcedAligner model. Unchecking both timestamp options releases it after 60 seconds unless it is kept loaded here. First load can add roughly 3.3 GiB of VRAM; some compiled CUDA context may remain cached after release.
Detecting aligner configuration…MCP connectivity
Disabled
Allow trusted AI agents to inspect deployment health, transcribe shared files, and use runtime controls. This is not authentication; keep the service on a trusted private network.
Endpoint and shared file directory are being detected…Realtime defaults
Persistent starting values for new browser and API realtime sessions. Individual API requests can still override them.
Readiness unavailable.