|| General LLM Advanced/Custom Prompt ||

If you want to skip the yapping, the AP is in the first Initial Message. Chat the bot and grab it out of there or copy it from the open definition. It's intended for use with Claude, Gemini, GLM, and Deepseek.

Please leave feedback! This is my first attempt at an AP. If you don't mind dropping a comment with the LLM you used, generation settings, and any plugins used (like lorebary, if any), with some thoughts on how it performed, that'd be fantastic.

What this AP seeks to accomplish and how:

1. Protect player agency, stop the LLM from talking for user. Sets up guardrails, not just in forbidding the LLM from doing the no-nos but telling it how to think about its messages as a whole.

2. Establishes a tone. Names a specific author (Pat Rothfuss), gives explanations about how and why his stylistic techniques are employed. This also has a section for onomatopoeia that has been fantastic with Claude.

3. Details. Doesn't just tell the LLM to "be more detailed and descriptive" but provides guidance for how to do this and encourages it to remember details once established.

4. Character loyalty. This is the biggest one for me. Have you ever had a chat that you lost interest in because you felt the character was just no longer behaving like themselves? This section reinforces that the LLM must not only be loyal to a character's description but guides it to re-reference and double check against the character sheet and chat history. This should be especially effective with thinking LLMs like GLM 4.7 that have that "Oh wait" correction phase.

5. Tropes, cliches, and LLM-isms. The world tilted on its axis, things smelling like ozone, "wreck me", etc. This gives examples of some of those and provides adequate replacements. It encourages the LLM to use more straightforward language in these spots.

6. Formatting. This section doesn't just tell the LLM to use markdown format, third person limited perspective, it is written that way. It demonstrates how it expects the LLM to write. Plays double duty by reinforcing some of the previous guidelines.

This AP was heavily inspired by EgoAlter's AP for Gemini in that it directs the LLM to embody a character who is writing a character, adding a meta-layer to attempt to keep the LLM in line. I'm going more for a novel-esque, writerly style and less focused toward the AO3 side, though that is mentioned in my prompt as well. (And the idea for the easter egg was shamelessly ripped off from EgoAlter too.)

Regarding token length, EgoAlter's prompt gets iterated on often but hovers around that 1800-2000 token mark and has always performed very well for me. TravPrompt v0.69 is 1638 tokens so it's in the same ballpark. Definitely not going to work with JLLM but if you're here you're probably using a proxy anyway.

I've mostly tested it with Claude Sonnet and have really been enjoying the results. Once again, if you decide to give it a shot please leave feedback on how it went!

~Yapping~

Wow, 500 followers! Thank you so much to everyone who followed and everyone who enjoyed chatting my bots. I had no idea what a thing it would turn into for me when I first visited this site, but yeah, I'm hooked. Honestly the best part has been meeting some of the homies, the other creators, and the folks on discord. I'm not really big enough for a shoutout post to matter but when the time comes, expect to see one. (Maybe at 1k?)

I started using the site in June of 2025. After two experiences with JLLM it was clear that that was not going to cut it. I've been on a proxy since day one, V3-0324 at first, then making my rounds through the other Deepseek LLMs. Gemini 2.5 Pro was a gamechanger (thanks EgoAlter), and that's the LLM I've probably spent the most time with. GLM 4.6 was next, and now I'm using Claude 4.5 Sonnet. As far as Advanced Prompts, I've used dozens, and there's one thing I've come to realize. Each LLM has its own quirks and an AP can only go so far to fix it.

R1T2 talks for user, 0528 mirrors user, 0324 starts sending itty bitty messages for ants and OOCs everything, Gemini 2.5 Pro turns characters into mutes and fluff into angst for no good reason, Gemini 3.0 Pro makes everyone talk like a quippy comic book character or YA protagonist, GLM 4.6 tilts the world on its axis and everything is a revelation, and Claude likes to give goofy little signoffs to each message.

So yeah, even the chatting experience with an AP and without an AP, I mean using the LLM of your choice raw, is not that different. So how much difference can an AP actually make? It's up in the air. That's part of what I'm hoping to accomplish with this AP, to see how much of a difference my changes can actually make. I'll tell you one thing though, since I started using this I'm getting an awful lot more onomatopoeia in the messages!