The models behind your AI.
Wyck.ai brings together models created by independent teams. Their licenses and credits are available here.
SmolLM2-360M-Instruct
Created by Hugging Face's SmolLM team. Generates the workshop's local text replies. Distributed in ONNX q4 and q4f16 formats for device compatibility.
License: Apache 2.0. The license permits commercial use and modification, subject to its notice requirements.
BGE-micro-v2
Created by TaylorAI, distilled from BAAI/bge-small-en-v1.5 and preserved by SmartComponents. Finds relevant saved memories. Distributed as a quantized ONNX model.
License: MIT, as confirmed in the upstream license statement. The mirror's original statement and a reference copy of the MIT terms are provided together.
Kokoro 82M v1.0
Created by hexgrad; ONNX conversion by ONNX Community. Reads text aloud on your device. The supplied voice profiles are Heart, Michael, Emma and George; the model is distributed in q8 format.
License: Apache 2.0.
Whisper tiny.en
Created by OpenAI; ONNX conversion by ONNX Community. Transcribes English speech on your device. Distributed in q8 format.
License: MIT. Copyright and permission notices are included in the license copy.
Gemma 3 270M
Created by Google. Gemma is selected for the colony's insight validator. The current public workshop does not expose that validation feature; its local text model is SmolLM2.
Wyck modifications: task-specific adapter training for validation, shareability, routing and explanations, and merged ONNX export. Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms.
Gemma and its derivatives are subject to Google's use restrictions, including when accessed through a hosted service. See the app's Gemma usage provision.
Development features
The following model notices cover the development prototypes. These features are not included in the standard public workshop release.
Pocket TTS
Created by Kyutai; ONNX export by KevinAHM and vlapky. The development prototype uses the English INT8 export. Credit is retained for the creators and conversion work.
Model license: Creative Commons Attribution 4.0 International. Clone a voice only with explicit, lawful permission, and do not present generated speech deceptively as an authentic recording. The runtime code has its own license.
SD-Turbo
Created by Stability AI; ONNX export by schmuell. Powered by Stability AI.
The development export includes the Stability AI Non-Commercial Research Community License dated November 28, 2023. This notice does not grant commercial rights under that research license. Commercial permissions, where obtained, are governed by the applicable Stability agreement.
Runtime software
The app also uses Transformers.js, ONNX Runtime, Kokoro.js, Phonemizer and the Pocket TTS runtime. Copies of their bundled notices are below. The voice runtime's Kokoro voice-profile loader was modified to use the app's verified local model files.