Third-Party Notices

CLAIR is built on open-source software and works with AI models made by others. We are grateful to their authors. Each component is licensed under its own terms, which apply to that component. The complete list of the components in each CLAIR release, with their versions, copyright notices and full license texts, is in the file THIRD-PARTY-NOTICES.txt in CLAIR’s installation folder. That file is generated from the release itself. Some packages contain files under more than one license; the file lists each one. Where a license gives you rights that the CLAIR Terms of Use restrict, that license controls for that component.

AI models

CLAIR does not include AI models. When you choose one, it is downloaded to your computer from the Ollama model registry, and CLAIR names the exact model and its license before the download starts. CLAIR does not modify model weights.

Built with Llama. Llama 3.1, Llama 3.2 and Llama 3.3 are licensed under the Llama 3.1, Llama 3.2 and Llama 3.3 Community Licenses, Copyright © Meta Platforms, Inc. All Rights Reserved. Use of Llama models is subject to the Llama Acceptable Use Policy.

Gemma 2 and Gemma 3 are provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms, including the Gemma Prohibited Use Policy. Gemma 4 is licensed by Google under the Apache License 2.0.

Qwen models offered in CLAIR are licensed by Alibaba Cloud under the Apache License 2.0. Phi models are licensed by Microsoft under the MIT License. Mistral models are licensed by Mistral AI under the Apache License 2.0. DeepSeek-R1 is licensed by DeepSeek under the MIT License; its 7B, 14B and 32B sizes are built on Qwen (Apache License 2.0) and its 70B size on Llama 3.3, which carries the Llama 3.3 Community License.

Components with special conditions

Component	License	What it means for you
GTK 3 runtime libraries (GTK, GLib, Pango, Cairo, HarfBuzz and related), used to create PDF reports	GNU LGPL 2.1 or later, with some parts under other licenses listed in THIRD-PARTY-NOTICES.txt	You may modify and replace these libraries, and get their source code (see below)
fpdf2	GNU LGPL 3.0	Same as above
pyphen	Mozilla Public License 1.1 (one of its license options, chosen by us)	You can get its source code (see below)
certifi, tqdm, orjson (some files), buffer-pipe, leb128	Mozilla Public License 2.0	You can get the source code of the covered files (see below)
Inter and JetBrains Mono fonts	SIL Open Font License 1.1	License text ships with the fonts
Ollama	MIT License	Copyright notice and license text ship with CLAIR
Python runtime	Python Software Foundation License	License text ships with CLAIR
Microsoft Visual C++ runtime and WebView2	Microsoft redistributable terms	Distributed under Microsoft’s terms
Machine-learning library pack (PyTorch, XGBoost, scikit-learn and others), downloaded on first use	BSD, Apache 2.0, MIT	Notices included in the pack and in THIRD-PARTY-NOTICES.txt
All other components	MIT, BSD, Apache 2.0, ISC and similar	Copyright notices, license texts and any Apache NOTICE files included in THIRD-PARTY-NOTICES.txt

Source code for LGPL and MPL components

The exact source code of every LGPL- and MPL-licensed component in each CLAIR release, including any changes we made, is available to download free of charge from the same place as the installer: downloads.clairanalytics.org/source/, in a folder named for the release. We keep each release’s source available for at least three years after we last distribute that release, and if CLAIR changes hands, the new owner takes on this commitment. You can also request a copy from [email protected].

The LGPL libraries are installed as separate files in CLAIR’s installation folder, and CLAIR loads them from there, so you can replace them with your own modified versions. You may also reverse engineer CLAIR as far as needed to debug such modifications.

Sample datasets

CLAIR’s sample datasets are for demonstration only. They are adapted from the UCI Machine Learning Repository and are used under the Creative Commons Attribution 4.0 International License. Each was changed for CLAIR: columns were renamed into plain English, coded values were turned into words, some columns and rows were removed, an ID column was added in most, and a sample of rows was taken.

• Superconductivity Data. Kam Hamidieh (2018). doi.org/10.24432/C53P47. Column names lowercased; 2,500 and 700-row samples of 21,263 rows; numbers rounded to 5 decimals.
• Student Performance. Paulo Cortez (2008). doi.org/10.24432/C5TG7T. Math and Portuguese tables combined with a course column; four columns removed; values in plain English; ID added.
• Predict Students’ Dropout and Academic Success. Valentim Realinho, Mónica Vieira Martins, Jorge Machado and Luís Baptista (2021). doi.org/10.24432/C5MC89. 25 of 37 columns kept; values in plain English; ID added.
• Diabetes 130-US Hospitals for Years 1999-2008. Beata Strack, Jonathan P. DeShazo, Chris Gennings, Juan L. Olmo, Sebastian Ventura, Krzysztof J. Cios and John N. Clore (2014). doi.org/10.24432/C5230J. Identifiers and sparse columns removed; two of 23 medication columns kept; codes grouped; primary diagnosis mapped to its ICD-9 chapter; outcome made “readmitted within 30 days”; rows with unknown gender or race removed; ID added.
• EEG Eye State. Oliver Roesler (2013). doi.org/10.24432/C57G7J. Columns renamed; 890 rows with electrode-disconnection spikes removed; ID added.

These are public research datasets that were published without names or other direct identifiers, and CLAIR removed the record ID columns that remained. That does not make them suitable models for real patient or student work.

Word list

CLAIR’s typo checker uses a list of four-letter English words from dwyl/english-words, released into the public domain under The Unlicense.