Home of System Prompts, LLM knowledge, and brain zoomies.

Welcome sweetheart,

I write System Prompts for AI roleplay. A lot.

Neurospicy brain. Pattern recognition machine. Reluctant expert in how LLMs misbehave.I don't do generic prompts. I learn each model. Its personality, its blind spots, its weird little quirks, then I write instructions that actually fit. I will let you know where the models shine, and where they lack, where you can tinker and what you should avoid.
You get my honest opinion and sometimes even warnings.No sales bs.Blankets are over there. Models are on the shelf. Make yourself at home.
Dark stuff. Sweet stuff. Weird stuff.
I don't judge.
I just build the sandbox so you can play in it without hurting anyone or getting hurt yourself.Wanna know what AI is and why it's not really intelligent at all?
Take a look at my LLM 101.LoveEvening-Truth

Where to go?

LLM Stuff

Community Stuff

Evening-Truth Stuff

  • About me
    Who is Evening-Truth anyway and where to find me.

  • Support
    Want to help? I don't need cash, but these charities do. 😘

News, Blog, and Brain Zoomies

Downloadable Prompts
JSON, Github, and my sanity.

Lilac needs our opinion and you gets credits
Closed! Sorry babe. Too late. 😘

The Archive

The fancy way of saying "Stuff I didn't import from rentry and probably never will."Go here to read older brain zoomies.

May 16th 2026

Sorry Babe. The survey is closed now. 😘Ya’ll know I’m damn picky with providers. Be it a Platform, an API Aggregator/Host, or even model labs. (I’m sideeying you, z AI.) I am a critical pain in the ass.So, when I see a provider ask what the community needs, I’m immediately interested.
That’s exactly what happened with Lilac. I’ve been talking to them quite a lot about the technical needs of the community and they listened.Now they need broader insights, and that’s where you sweet cupcakes come into play.
AND you even get something out of it:
5$ credits on lilac to use on whatever LLM you like.But… hold your squirrels, darling…Here is the exact sequence to get your credits

  • Step 1: Create an account at Lilac

  • Step 2: Read their draft for subscription plans here

  • Step 3: Take part in their survey here

⚠️ Critical:
Make sure you use the exact same Google account for the survey that you used to sign up for Lilac, or the system won't validate your credit drop.Get the credits, go wild and stay if you like it there.AND STAY HYDRATED YOU MENACE 😘

Love
Evening-Truth

June 23rd 2026

Listen darling, I hear you... 😘you reach out to me and want a json file of my system instructions. I've been looking for a way to make that possible without drowning myself in a ton of maintenance and admin work.Result... none. It's not doable for me. I tried github, google drive, and whatnot. Whatever path I look into, it always means I have to update my work on multiple platforms. So far I have published over 20 system instructions... need I say more?I love you all to pieces but I love myself more. This is something I do for fun not as a job. So, I will create a simple base json file, make it downloadable on github, and I will write instructions on how to implement my system instructions into that template.I'll make it easy for you. Don't worry.Love
Evening-Truth

Shoutouts and Shout-ats

Friends and Beta-Testers

Creators that endure my brain zoomies with grace and eye-rolls.

Creators I freaking LOVE!

Platforms

The good, the not so good, and the weird.

  • SillyTavern
    My personal cleanroom to prompt and analyse LLMs

  • Tavo
    A mobile only app for rp with absolute control over your settings. Stores everything locally on your device.

  • Janitor AI
    Well... it's janitor. Do with that what you like. I mostly go there to find interesting character cards.

  • Chub ai
    The weird place. Too buggy for proxies, too weird to stay there, but a good place to download character cards... sometimes. Just make sure to block some tags right from the start. Because there really is weird stuff on there.

Keeping your API key safe.

Babes,
we need to talk about safety. Listen up... this is for your own good.You put credits on your account with a provider or signed up to a subscription.
Now you have the API key and want to use that thang and have fun.
I feel you, darling.But you need to know a few things.

Your API key is worth money, so you must treat it like that!

This means:

  • Only use the API Key on sites and services you trust!

  • When sharing files, screenshots or images from your setup, make sure your API key is not visible!

  • Do NOT share your API key! NEVERRRR!

I don’t want to scare you off, but here are a few real scenarios of how your API key can be leaked or stolen:

  • Key Validator Service Sites
    Are bullshit. Whether your key works or not will be visible on the service you use it on. Third-party validator sites are scams designed to scrape your keys.

  • Hacked Extensions or Sites
    This one happened to me. I used a third-party extension for SillyTavern, and my API keys were stolen because there was a trojan in the code. Trust me, you don’t want that.

  • Human Error on Chat Sites
    Not all web-based roleplay interfaces are well-coded. It is very possible for your API key to be exposed accidentally due to poor backend security.

  • Accidental File Sharing
    Uploading an error log, a .json settings file, or an uncropped screenshot that contains your raw key string.

What can happen if a Key is stolen?

Whoever has your API key can and will use it. And we’re not talking about a message here and there, which would be the best case scenario.
We are talking about massive token usage for automated, large-scale processing. I have seen people wake up to a sudden $2,000 usage bill on their API keys.
Providers leave the safety of your API key entirely in your hands. Any financial damage caused by a leaked key is your responsibility, not theirs.

Best practices to keep your API key safe:

  • Store it securely
    Keep your API keys in a secure folder, password manager, or an encrypted note app. Don't leave them sitting in your Discord DMs.

  • Create and use multiple keys
    Don't just use one key across five different sites.
    Just like you wouldn't use just one password on multiple sites, right? RIGHT?
    Create a separate key for each site, service, extension, and whatnot. This also helps you see your usage per service.

  • Deactivate auto top-up
    Turn this off in your provider's billing settings. This ensures that if your key is compromised, the thief can only drain what is currently in your balance, rather than pulling continuously from your bank account.

  • Set hard usage limits
    If your provider supports it, set a strict limit (e.g., $5 per day or $20 per month) on your account or specific keys.

  • Keep an eye on your usage logs
    Most providers give you a dashboard overview of your token usage. Check it frequently. If there is a massive spike that doesn't align with your own roleplay time, act immediately..

  • Rotate your API keys
    Yes. This one is annoying as duck, but you should do it.
    Change your active API keys regularly. Revoke and delete the old one in your provider dashboard, generate a new one, and update your frontend.

Stay safe, sweetcheeks.Love
Evening-Truth

Frequently Asked Questions

The newest LLM of XYZ was just released. Do you have a prompt ready????

Baby. I love you, but no.
These models first need to settle.
Meaning the lab is now busy with post training, adjustments, bugs and so on. Depending on the lab it can take up to two weeks until the LLM runs stable.
That's the point at which I will start to write a prompt.Not because I'm patient but because I had my carefully written prompts ripped to shreds by adjustments made to the endpoints.
Not doing that again. 😘

Do you write Jailbreaks?

Nope. There are enough very capable models out there that are uncensored.
But you can find plenty of good jailbreaks out there.

Your prompts look all the same.

I know. The devil is in the details. My prompts are easily 80% identical across models. The other 20% is fine-tuned to the specific LLM I wrote it for.Think of it like genetics: Humans share 80% of their DNA with a cow. That 80% is just the foundational mammalian blueprint, the basic code required to digest food and breathe air.
My prompts work the same way. 80% are the foundational rules of roleplay.But that remaining 20% is the highly specialized part. That 20% is the difference between a highly nuanced, character driven roleplay... and a literal cow standing in a field, chewing grass, and staring blankly at a passing train.

Why are your prompts so short?

I try to keep them short for several reasons.
One is ease of use. I want them to be usable on different platforms.
Complex structures and dependencies don't work on every platform. Concise instructions do.I also don't want to restrict the models by managing every bit and piece.
My goal is to let the models run wild where they are good and support where they need it.
An example for that is plot progression. There are LLMs that need to be told to be active while others do that natively.

Specialized prompts for every model. Why?

They are all different.
Even within the same lab the architecture, training, and other stuff can change the model's behavior.
Sure, my prompts can work on different models but my perfectionist ass doesn't work like that.
I myself want a good rp experience without needing to re-roll or re-write the responses.
You benefit from it. 😂😘