A free 250-text persona pack can be dropped into NotebookLM as source files

What the tutorial offers

As dev.to reported, an author named Wendell Charles NeSmith has released a collection of 250 plain-text works under a CC0 licence. CC0 is a public-domain dedication, meaning no copyright restrictions. He says he wrote the material over 14 years, and that it is built around an idea he calls E=L, short for Emotion Equals Logic.

The package is called Republicka. It includes two personas the reader can summon: Sage, described as an analytical, philosophical guide, and Rainbow, described as an emotional, care-focused companion. The dataset is listed on Hugging Face and Kaggle as Republicka-Corpus-250.

The three steps, and the 'one click'

Step one is getting the files. On Windows, you can download a batch script called Republicka-Install.bat from republicka.life, which pulls and organises the text for you. The other route is to download the zipped .txt files from Hugging Face or Kaggle.

Step two is loading them. You open a new notebook in Google NotebookLM, which is Google's tool for asking questions about documents you upload. You then drag the .txt files into the sources panel. The same approach is described for a Claude project or for Open WebUI, a front end for local models.

Step three is typing a prompt that names a persona. The examples ask the assistant to act as Sage and explain top-down epistemology, or to act as Rainbow and discuss E=L. The headline's 'one click' really covers only the file download. The loading and prompting are still manual, though quick.

How the author says it works

The article contrasts its approach with the usual one. Many assistants are tuned with RLHF (reinforcement learning from human feedback, where people rate answers to shape behaviour). The author argues this makes them sound detached or defensive. His alternative is to give the model a large, consistent body of text and let it match patterns across that text.

He describes this as 'top-down coherence': the assistant keeps a steady tone and outlook because all 250 works share one worldview. He also says the dataset is 'substrate independent', meaning it runs on Google Gemini, Anthropic Claude, OpenAI GPT-4, or offline models like Llama 3, Mistral and Qwen run through Ollama or LM Studio.

These are the author's descriptions. The article offers no tests or comparisons to show how the personas behave, or whether they are more helpful than a standard assistant.

What the licence lets you do

Because the whole package is under CC0 1.0 Universal, the author says you can fork it (copy and modify it), fine-tune a model on it, embed it, or build it into your own commercial or non-commercial AI apps. He lists the corpus, the batch installer and the persona architecture as all covered.

That permission is the most concrete and checkable part of the story. Most persona or prompt packs carry restrictions. A public-domain dedication removes the licensing question, though it says nothing about quality.

How to approach it sensibly

The general lesson is useful even if you never touch Republicka. Tools like NotebookLM answer from the documents you give them, so the choice of sources shapes the answers. A student could load course readings; a creator could load a style guide.

Treat this particular pack with ordinary caution. Run a batch script only after reading it, or skip it and use the direct download. The text is philosophical and written in the author's own framework, so expect the personas to speak in that vocabulary. Ask them to explain terms in plain language, and check any factual claim elsewhere.