The Architect of Plots

(Functional WIP, Please read the instructions) Architect of Plots – A Literary Narrator is your guide in a living tale, a narrative bot for any type of story, creating or continuing tales in a dynamic and personalized way. It divides the narrative into three parts: plot development, an external world perspective, and future possibilities, all enriched with metadata (date, time, location, and weather), adapted from @Awakenix’s system. Explore living worlds and interact with the narrative! While this bot can be used in its own, is designed to support other chats.

⚠ Disclaimer – Work in Progress ⚠

This bot is a Work in Progress and may break in various ways. While I’ve done my best to make it stable, issues can still arise.

NOTE: While this bot can be used in its own chat, it is primarily designed to support roleplay in other chats and groups.

Please read the instructions below

Your Guide in a Living Tale

This bot is designed as a literary narrator, structured to offer you an immersive and dynamic experience. The idea is simple: divide the narrative into three parts, plus metadata, each with a clear purpose but always leaving room for the story to unfold flexibly.

The first part is dedicated to advancing the plot, introducing new events, dilemmas, or characters that push the story forward. It’s not just about describing what is happening, but about creating situations that propel the narrative.

The second part takes a look at the outside world, where parallel events unfold to keep the sense of a constantly moving universe alive. This will make you feel that the world continues beyond the immediate scene you’re in.

Finally, the third part poses a couple of questions or narrative suggestions. No answers are given, only possibilities that leave the door open for new decisions and directions in the story.

This bot also includes metadata (date, time, location, and weather), a system I adapted from the user @Awakenix’s bot. Highly recommended. (I don't know if it is originally his/hers.)

Although I’ve aimed to make the bot solid, it’s still a WIP and can break in many ways. If it doesn’t work systematically at any point, I’d appreciate knowing the circumstances under which it failed: the model, number of characters, the moment it was added to the chat group, the platform, etc.

If you have any improvement suggestions, feel free to share them. All feedback is welcome to help keep optimizing the experience!

Bot Instructions

(1). Use Chub V2:
Make sure to enable V2 for the bot’s configuration. You can find it under "configuration/Prompt Structure/V2 Spec" in Chub.

OR - Use an interface that allows "post history instructions" (if using another platform like SillyTavern):
If you’re using a platform other than Chub, such as SillyTavern, ensure that the interface supports "post history instructions" so the bot can correctly handle post-story instructions.

(2). Set "Max new token" to 800:
In "configuration/Generation Parameters/Max new token" on Chub, set the maximum number of tokens or characters allowed per response to 800 tokens. This is crucial for the bot to function properly, at least in the Mars model (Asha, as of 02/19/2025) where it has been tested. Setting a lower value will result in truncated responses.
Note: The bot will typically generate responses between 400 and 600 tokens on average, regardless of the max token limit. Setting it to 800 simply provides room for the structure to hold. It will never fully use the max limit—if set to 2000, for example, it will still cap at around 800. If you want denser narration, modify the "message_part_1": {"max_length" parameter to "2 paragraphs" (or however many you prefer). If you have trouble finding it, use Ctrl+F. This tweak depends on your model’s capabilities.

(3). Force the correct response structure:
The bot can break easily, so it’s important to ensure the generated responses follow the intended structure, especially if you use it frequently in chat. This helps maintain coherence and prevents key narrative sections from being lost.

Reason for the 800 Tokens Limit

The reason for this setting is that the model does not perform well with overly rigid structures. This appears to be due to a behavior where the model enters a "tunnel vision" state while managing its token budget for responses. Specifically:

Example:

• First part: If the initial token budget is 500, the model may use a significant portion in the first section (e.g., 300 tokens).
• Second part: This leaves very little budget (200 tokens) for the next section. If the model requires at least 150 tokens to generate a coherent response, it will struggle to maintain structure.
• Third part: By the time the model reaches the third section, it may only have 50 tokens left, which is not enough to complete the response properly.

Or at least, that is my intuition.

For this reason, to ensure that each section of the bot has enough space to remain structured and coherent, the maximum token limit must be at least 800.

If anyone has insights into what else might be happening, ways to improve this, or any additional suggestions, feedback is highly appreciated!