> For the complete documentation index, see [llms.txt](https://novatore.gitbook.io/novatore-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://novatore.gitbook.io/novatore-docs/novatores-system-prompts.md).

# Novatore's System Prompts

*<mark style="color:$info;">Janitor:</mark>* [*<mark style="color:$info;">@novatore</mark>*](https://janitorai.com/profiles/fec6db1d-aff4-4beb-a8c0-d0b0ae151b70_profile-of-novatore)&#x20;

This Gitbook houses System Prompts and Generation Settings for my logic-heavy RP bots.

These prompts are tailored for my bots, designed to force the LLM models to read my bots' logic gates as executable code payloads. Meanwhile, the writing style is paced and styled like a comicbook/manga.

#### GLM vs. DeepSeek

The things I've noticed between these models.

* GLM is a method actor. It's really good at portraying characters and tropes. Especially good at slow-burn genres.&#x20;
* DeepSeek is good at maintaining logical rules. It follows prompts to a T. Though I don't really like how it portrays characters.

Note (1): GLM requires "Warm Tokens" and narrative justification to follow my bot's rules. This is why it uses more tokens than DeepSeek for its prompt.

Note (2): I have a raging boner for GLM. Specifically, GLM-4.6. So that's the most optimized prompt I've made and tested.&#x20;

Update 2026/05/24: **Kimi K2.5/2.6 Thinking** is my new best friend. DS prompts are removed because they're not optimized.&#x20;

| Navigation                                                                                                                         |
| ---------------------------------------------------------------------------------------------------------------------------------- |
| [*GLM Prompts for v5.1 and v4.6*](https://novatore.gitbook.io/novatore-docs/pages/v95WLjdxYeBCs7yF1sZs#glm-5.1-prompt-1271-tokens) |
| [*Kimi K2 Prompt*](/novatore-docs/prompts/kimi.md)                                                                                 |

{% hint style="info" %}
*How the fuck did you get to use DS v4 and GLM v5.1 so early?*\
\
NanoGPT. 8 dollars a month. 60M token quota per week. If you're using Janitor like 1-3 hours a day, you'll have all the tokens you need.&#x20;
{% endhint %}

{% hint style="info" %}
*But Chutes and OpenRouter—*\
\
*Shush*. Chutes lobotomizes their models and OR has shitty providers. Trust me. Use NanoGPT.
{% endhint %}