ciao, bellas! 🌹

a step-by-step proxy tutorial
for nano-gpt & janitorai.

________________________

if you follow me, you know i test most of my bots (and make them) with the use of proxies in mind. JLLM is definitely improving (or on its way to improving), but as we all know, it's still on beta. to get the MOST out of my bots (or any others, really), i definitely recommend using a proxy. now, lots of people have told me they don't even know what that means.

first and foremost: most good proxies are NOT free. so let's get the cat out of the bag right away. there ARE some good free proxies out there, but they're usually free only for a time for testing purposes and quickly get taken down after enough people discover them.

there are different websites through which you can get your proxies. for this tutorial, i'll talk about the one i'm using, which is nano-gpt.

so let's get started.

one. what is a proxy?

think of a proxy as a middleman between you and the actual AI model. when you use JanitorAI normally with JLLM, you're using JanitorAI's own language model. when you use a proxy, you're routing your conversation through an external service (nano-gpt) that connects you to different (and often more powerful) AI models like GLM 5, Kimi 2.5, or others.

why does this matter? better models typically mean:

• smarter responses that stay in character

• better memory and consistency throughout long roleplays

• less filtering (depending on the model you choose)

• more creative and nuanced writing

the proxy handles the connection, and JanitorAI just sees it as another API endpoint to send requests to.

two. how much does nano-gpt cost?

nano-gpt offers a few different options:

• [ tier - cost - what you get ]

• [ pay as you go (PAYG) - you decide - add money to your nano-gpt wallet/balance and use any models you want. each attempt at a generated message has a cost based on tokens, lorebooks, model you use, etc. i don't recommend this unless you are more aware of what everything does and costs.]

• [ pro -12 USD/month - 60 million included input tokens per week, access to most models] (this is the one i use & recommend!)
  Edit: It was 8USD before, recently changed to 12 USD, so I updated this.

the free tier is great for testing things out and seeing if you like how it works. if you're going to be roleplaying regularly though, i'd recommend at least the pro tier so you don't hit message limits in the middle of a good story. also, models that i personally recommend such as GLM 5 and Kimi 2.5 are only accessible with a pro account.

KEEP IN MIND: there are models such as Claude, TEE models, GLM 5.1 that are NOT included with pro due to being MORE expensive than normal. they're better, obviously, but you can only use them via nano-gpt if you add money to your nano-gpt balance (you don't need a subscription, you can use this option for all of it if you want, but having a subscription gives you a discount).

three. create your nano-gpt account

• go to nano-gpt.com

• click the arrow to expand the menu on the right and look for the "sign in" button in the bottom corner

• enter your information and create your account

• you may need to verify your email - check your inbox for a confirmation link

• once verified, log in to your account

four. subscribe to nano-gpt pro (optional but recommended)

if you want to avoid strict message limits and access the better models:

if you want, you can use my referral code! doesn't change anything for you, but gives me a discount, i think!

• look for a "subscription" button on the site

• select the plan that works for you

• enter your payment information

• confirm your subscription

once subscribed, your account will show your new plan status. you can now access the fuller range of models and enjoy higher token caps.

five. open a notes app and copy the following information to it

before you configure JanitorAI, you need three pieces of information saved somewhere handy. open your notes app (notepad, apple notes, word, whatever you prefer that is PRIVATE) and copy these:

• your api key

  • found in your nano-gpt side menu, usually under "Account" in "API"

  • when you enter that page, you'll see an AutoGenerated key. don't use that. create a new one, name it "JanitorAI" or whatever else you want and copy that to your notes app

  • it's a long string of random characters

  • never share this with ANYONE under ANY circumstances - this is your personal key and anyone with it can use your tokens. treat it like a password.
    

• proxy url (copy the link below and paste it to your notes app)

  • https://nano-gpt.com/api/v1/chat/completions
    

• your chosen model

  • you'll pick this in the next step, but save it here once you decide

six. choose a model

on the menu on the right, click the button that says "models". it'll take you to a new screen and, there, you'll select "text" on the tabs on top. if you see a huge "text models leaderboard", just close that by clicking the arrow pointing up on the corner (^). scroll down and choose your model.

once you scroll down, you'll see this:

'show subscription-included models' should always be toggled on. save yourself the headache of choosing a model that is not included in the subscription (unless you're going the PAYG route).

MODELS I PERSONALLY LOVE & RECOMMEND:

included in pro tier

• GLM 5 Thinking / (zai-org/glm-5:thinking)

• Kimi K2.5 Thinking / (moonshotai/kimi-k2.5:thinking)

• GLM 4.7 Thinking / (zai-org/glm-4.7:thinking)

• GLM 4.6 Thinking / (z-ai/glm-4.6:thinking)

• DeepSeek R1 0528 / (deepseek-ai/DeepSeek-R1-0528)

the list above is structured like this:

• Name / (model)

the thing between parenthesis is what you need to save in your notes app in the "model" part. without the parenthesis, by the way.

seven. add your proxy information to janitor

1. click on any of your chats and open it.

2. click the "janitor" button on top of your screen and your API Settings will appear.

3. select "Proxy"

4. add your custom prompt to the Global Prompt window (here is mine, if you want).

5. click the '+ New' button and add the information you saved to your notes app here. it should look like this:

keep in mind that you can have many proxy configurations, but you need separate configurations for each model. if you want to switch between models to test them out, just make new ones. like this:

every time you switch between models/configurations, you have to REFRESH YOUR PAGE, otherwise you'll get an error. every time you SAVE an API, delete it, modify it, modify your settings or your prompt, save and refresh!

eight. tweak your settings

now, this is more personal. it's trial and error and there are MANY creators out there that can give you their preferred settings for proxies. for me, this is what works:

• show thinking: off. - thinking will still be there and you'll be able to see it if you click to edit the bot's reply. it'll be hidden between the <think>tags</think>. it's extremely useful to see if your custom prompt needs tweaking or, in my case, to see what's the AI's train of thought with some of my bots so i can tweak them before i post them.

• temperature: 0.9 - this is the sweet spot for me. anything above it and the LLM starts adding chinese symbols & hallucinating like nobody's business. sometimes i still get some chinese characters lmao, but it's rare and it doesn't bother me.

• max tokens: 0 - if you want to limit your bots so they don't write a huge essay (could never be me), put that in your CUSTOM PROMPT. add the constraint that you want a max of three paragraphs, for example. don't limit your bot in the settings because sometimes it just makes the bot not finish sentences.

• context size: 31k - i oscillate between 28k and 32k at any given point. if i start noticing that the AI is hallucinating too much, making too many mistakes, changing the context size even by 1k (up or down) seems to give it a nice shake of 'WAKE TF UP'.

• advanced: yeah, i know, so many people have their own preferred settings for these, but with proxy i usually don't touch it. this is, like i said, trial and error. find other creators, see what their settings are, find what you like.

nine. save your info & settings and refresh the page!

there, you're set! start writing! <3 if everything worked, you should be good to go.

now, i'm not a proxy GENIUS, but if you encounter any problems, i can try and help in the comments. there are also other websites with different prices and different formats. this is just what's more affordable FOR ME.

if you want other proxy/model recommendations, go on janitor's reddit or even the official discord, i'm sure there's a megathread somewhere with people's recommendations and experiences!

Have fun! 💕