Shendru includes a fully local transcription pipeline powered by Whisper and ONNX Sherpa for speech recognition and speaker diarization. While this works entirely on-device, the output can occasionally feel less polished depending on audio quality, accents, or overlapping speakers.

For users who want higher-quality transcription and more accurate speaker detection, Shendru also supports connecting your own Deepgram API key.

You can create a free account at deepgram.com, which currently includes $200 in free credits — enough for roughly 400 hours of audio processing, covering several months of meeting transcription for most users.

When using Deepgram, Shendru can also generate meeting summaries directly through Deepgram, meaning you do not need a separate OpenAI API key for summarization.