Roder
Capabilities

Speech

Roder exposes a provider-neutral speech-to-text service. Clients own microphone capture and encoding; Roder accepts bounded audio payloads and normalizes provider results into text, language, duration, segments, and metadata.

Providers

  • openai-speech supports OpenAI transcription models including gpt-4o-mini-transcribe and whisper-1.
  • google-speech supports Google Speech-to-Text with access tokens, API keys, or authorized-user ADC.
roder speech providers
roder speech transcribe clip.wav --provider openai-speech --model gpt-4o-mini-transcribe
cat clip.wav | roder speech transcribe - --format json
roder speech transcribe clip.wav --to-thread <thread-id>

Raw audio is not stored in thread transcripts. API transcript recording redacts audio bytes and sensitive headers before writing JSONL. Sending a transcript into a turn is explicit via --to-thread; audio never starts a turn silently.