Cascade

Worker lingua-cascade. One spoken turn through OpenRouter on https://inference.archangl.ai. IPv4: https://lingua-cascade.archangl.workers.dev.

Turn

POST /turn. Body is pcm16 at 16 kHz (16000). Send content-type: audio/pcm and x-lingua-sample-rate: 16000. A 200 returns pcm16 with x-lingua-transcript, x-lingua-reply, and x-lingua-sample-rate: 16000.

Caller

Worker lingua-voice reaches this engine through a Cloudflare service binding named CASCADE. Dummy host on that Request is cascade: POST https://cascade/turn. Harvest hang-up stays on worker lingua.

Bind

Gateway https://inference.archangl.ai. Auth is the gateway token in cf-aig-authorization, with BYOK alias or-DSH-MAIN. No AI binding: BYOK gateways answer that path with AiGatewayError 2001 for every id. Secrets: AI_GATEWAY_TOKEN, AI_GATEWAY_BYOK_ALIAS.

Models

STT

nvidia/nemotron-3.5-asr-streaming-multilingual-0.6b

LLM

nvidia/nemotron-3-super-120b-a12b:free

TTS

deepgram/flux-tts:free · flux-alexis-en