Levers AI: chat + voice

levers-sonic (LLM) levers-tashkeel (TTS)

Settings

Preset (auto-fills system prompt + TTS instructions; you can still edit after)

Sonic — LLM

System prompt

Temperature

Top-p

Max tokens

Tashkeel — TTS

Voice

Speed

Voice = ref name on the TTS server (today only default is meaningful). Speed multiplies the generation rate.

CFG value (1.5–3.0; lower = more expressive, higher = tighter to ref)

Reference audio override (URL or data:audio/wav;base64,...)

…or upload a WAV/MP3 (becomes a data: URI inline)

No file selected. Min ~3s, max ~30s clear speech.

assistant

السلام عليكم! كيف أقدر أساعدك اليوم؟

auto-send on EOU

tokens stream live; each sentence is voiced as soon as it completes