Introduction

sonix is the low cost voice API for production speech. It speaks the two TTS wire formats you already use, OpenAI and ElevenLabs, over CPU-first inference, so there are no GPUs in the price.

The low cost is structural, not subsidized: we rebuilt the stack around our own small inference engine instead of wrapping a general-purpose GPU framework. It serves compact speech models on plain CPUs, so synthesis costs a fraction of GPU inference and the price follows.

Point your existing SDK or HTTP client at your sonix base URL and keep the same request shapes:

Both surfaces share the same API keys. The OpenAI surface selects any deployed speech model; the ElevenLabs surface uses the one your deployment configures. The API also serves speech to text via /v1/audio/transcriptions and /v1/audio/translations.

Examples in these docs use $SONIX_URL as the base URL and $SONIX_API_KEY as the key. Get access to obtain a key, then start with the quickstart.