ASPArROW AI

Live TTS + LLM inference for the ASPArROW auction platform. CPU-hosted on VPS2 — no GPU, no usage fees.

Endpoints (bearer token required)

POST /v1/chat/completions — OpenAI-compatible chat (qwen3-1.7b)
POST /v1/audio/speech — OpenAI-compatible TTS, returns WAV
GET /v1/models · GET /v1/voices · GET /healthz

Voices

vivian · maya (female) — ryan · aiden · leo (male)
Piper neural voices, 22kHz. Voice cloning arrives with the Modal GPU tier.

Notes

The chat model is a small local model — good for auction scripts, lot descriptions and show rundowns; not a frontier model. For heavier work the MCP tools will prefer the Modal GPU tier once provisioned. Nothing here is faked: errors are explicit.