> ## Documentation Index
> Fetch the complete documentation index at: https://docs.withhopper.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Hopper

> Open-source voice models served for real-time latency behind an OpenAI-compatible API

Hopper serves open-source models for real-time voice agents: Google's Gemma 4 31B for chat completions, k2-fsa's OmniVoice for text to speech, and NVIDIA's Nemotron ASR for speech to text. All three run behind one OpenAI-compatible API at `https://api.withhopper.com/v1`.

<CardGroup cols={3}>
  <Card title="LLM" icon="message-lines" href="/llm/chat-completions">
    Chat completions on `gemma-4-31b`: streaming, tool calling, 25,000-token context.
  </Card>

  <Card title="Text to Speech" icon="waveform-lines" href="/tts/generate-speech">
    24 kHz streaming audio from `omnivoice`, with instant and pro voice cloning.
  </Card>

  <Card title="Speech to Text" icon="microphone" href="/stt/transcription">
    Batch and streaming transcription on `nemotron-asr`: word timestamps, keyterm biasing, turn detection.
  </Card>
</CardGroup>

<CardGroup cols={3}>
  <Card title="Quickstart" icon="bolt" href="/quickstart">
    Get a key, make a request, stream a response.
  </Card>

  <Card title="Models" icon="table" href="/models">
    The catalog: model IDs, context windows, sample rates, pricing.
  </Card>

  <Card title="API Reference" icon="code" href="/api-reference">
    Endpoint-level reference: auth, parameters, error shapes.
  </Card>
</CardGroup>