Skip to main content

​
Common Issues

  • No API key — Set OPENAI_API_KEY (for OpenAI Realtime) or GEMINI_API_KEY / GOOGLE_API_KEY (for Gemini Live). The proxy logs no OPENAI_API_KEY or no GEMINI_API_KEY if missing.
  • Network / firewall — Both providers use outbound WebSocket (wss://). Ensure ports 443 are open. The proxy retries 3 times with exponential backoff before failing.
  • Expired or invalid key — A 401 from the provider appears as Realtime API error in logs. Rotate the key and restart.
  • Wrong model name — Default models are gpt-4o-realtime-preview-2024-12-17 (OpenAI) and gemini-2.0-flash-live-001 (Gemini). Override with FERAL_GEMINI_LIVE_MODEL env var.
  • Checking logs — Voice logs under feral.voice.openai and feral.voice.gemini. Set FERAL_LOG_LEVEL=DEBUG to see per-chunk latency and connection attempts.