CallMe

The on-device models the CallMe AI iPhone app downloads with Download All Three (Settings → Models). Each folder is one component of the voice stack. The app fetches only the files it needs, checks each large file against its SHA-256, and resumes interrupted downloads.

Folder	Component	Model	Licence
`stt/`	Speech recognition	Moonshine Small Streaming (English), ONNX Runtime	MIT
`llm/`	Conversation	Gemma-4-E2B, MLX 4-bit	Apache-2.0
`tts/`	Voice	Ming-3.0-tts-0.5B, MLX Swift port, 8-bit (backbone, aggregator and DiT Linears; VAE decoder bf16)	MIT

All three are permissively licensed (MIT / Apache-2.0); see each upstream model card (Gemma 4: model card).

Everything runs on the phone; nothing is sent to a server during calls.

Downloads last month: -; Downloads are not tracked for this model. How to track

MLX

Hardware compatibility

Quantized

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support