moonlauncher/moon-models

Moon Launcher models

The on-device AI models Moon Launcher (Harbor Software) downloads and keeps up to date. Everything runs on the phone; nothing a person searches, says or receives leaves it.

models.json lists the newest version of every model the app uses and where to get it. The app reads it about twice a day and updates on Wi-Fi by itself.

What's here

Unchanged copies of Google's Gemma files, which otherwise need a Hugging Face sign-in:

File	From	Used for
embeddinggemma/embeddinggemma-300M_seq256_mixed-precision.qualcomm.sm8750.tflite	litert-community/embeddinggemma-300m	Search by meaning, on the Snapdragon 8 Elite's NPU
embeddinggemma/embeddinggemma-300M_seq256_mixed-precision.tflite	same	Search by meaning, on other phones
embeddinggemma/sentencepiece.model	same	EmbeddingGemma's tokenizer
gemma/gemma3-1b-it-int4.litertlm	litert-community/Gemma3-1B-IT	The assistant on phones without a Snapdragon
gemma/gemma3-270m-it-q8.litertlm	litert-community/gemma-3-270m-it	The same, on smaller phones

Open models (Whisper from OpenAI and Qualcomm AI Hub, and the GGUF builds of Gemma from unsloth) are not copied: models.json points to where their makers publish them.

Terms

Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms. A copy is in GEMMA_TERMS.md. Use of these files is subject to the use restrictions in Section 3.2 of those terms, including Google's Gemma Prohibited Use Policy.