Standing on tall shoulders

AI models on the stick

• Qwen 3.5 (4B and 9B) — chat models by Alibaba Cloud, Apache 2.0.
• nomic-embed-text — semantic search embeddings by Nomic AI, Apache 2.0.
• Kokoro — local text-to-speech voices, Apache 2.0.
• Whisper (small.en, via faster-whisper and CTranslate2) — local transcription, MIT.

Software

• Ollama — local model runtime, MIT.
• Node.js — application runtime, MIT.
• Tesseract OCR — reading scanned pages, Apache 2.0.
• ImageMagick — image processing, ImageMagick License.
• PyTorch — powers the local voice models, BSD-3-Clause.

Components under the GNU GPL

These are licensed under the GNU General Public License, which gives you the right to a copy of their source code:

• poppler — turns PDF pages into images so text can be read out of them, GPL-2.0-or-later.
• espeak-ng — pronounces English words that are not in the speech dictionary, GPL-3.0.
• phonemizer — converts written text into speech sounds, GPL-3.0.
• num2words — reads numbers aloud as words, LGPL.

Written offer: for three years from the date you received your Think Stick, we will give you the complete corresponding machine-readable source code for the GPL- and LGPL-licensed components above, for no more than our cost of distribution. Email thinkstickai@gmail.com and include your Think Stick version. This offer covers only those components — it does not cover Think Stick's own source code, which is not open source.

The full credits, versions, and license texts ship on every stick: open THIRD-PARTY-NOTICES.txt at the top of the drive, with the complete licenses in the licenses folder beside it.

If you maintain one of these projects: thank you — you made a private AI for regular people possible.