Open-source and AI model notices
Cosmo is built with open-source software and openly licensed AI models. The full texts below are the same files the installer places beside the app.
Cosmo
Cosmo © 2026 hackerlumu — Cosmo licence. Cosmo is derived in part from Coucou by Louis Raillé (MIT); that notice is included in the third-party notices.
Third-party software notices (full text)
Website
This website uses the Inter typeface by The Inter Project Authors, under the SIL Open Font License 1.1.
Local AI models (optional downloads)
None of these is included in the installer; each is downloaded only when you choose it, from its publisher's Hugging Face page, and checked with SHA-256. “Mini”, “Lite”… are Cosmo's names for these download options; the models were made by their creators, not by Cosmo.
| In Cosmo | Model | Creator | Licence |
|---|---|---|---|
| Mini | SmolLM2-135M-Instruct (GGUF by bartowski) | Hugging Face (SmolLM2 team) | Apache-2.0 |
| Lite | SmolLM2-360M-Instruct | Hugging Face (SmolLM2 team) | Apache-2.0 |
| Smart | Qwen3-0.6B | Qwen team (Alibaba Cloud) | Apache-2.0 |
| Pro | Qwen3-1.7B | Qwen team (Alibaba Cloud) | Apache-2.0 |
| Ultra | Qwen3-4B | Qwen team (Alibaba Cloud) | Apache-2.0 |
Run with llama.cpp (MIT), installed with Cosmo. Apache License 2.0 (full text)
Voice
| Component | Source | Licence |
|---|---|---|
| Speech recognition models (optional) | Whisper by OpenAI, ggml conversion by the whisper.cpp project | MIT |
| Speech recognition runtime | whisper.cpp | MIT |
| Natural Voice (optional) | Kokoro-82M v1.0 by hexgrad, ONNX export by onnx-community; six of its voices | Apache-2.0 |
| Pronunciation dictionaries | hexgrad/misaki | Apache-2.0 |
| Voice runtime | ONNX Runtime | MIT |
| “Hey Cosmo” detector | Cosmo's own small model, on two Apache-2.0 feature models (openWakeWord's mel spectrogram, Google's speech_embedding) | Apache-2.0 (feature models) |
| System voices | The voices installed in Windows | Windows licence |