DeepSeek Guide (+ Advanced Prompts) | DeepSeek Tutorial

DeepSeek Guide (+ Advanced Prompts) | DeepSeek Tutorial

19.8k

69.8k

by:@GoldAnnie

This is NOT a roleplay bot. It contains info about DeepSeek LLM and how to achieve a dramatically better roleplay experience with it. You'll find general information, extensive comparisons, a tutorial and several links to other guides. There's a table of contents in the beginning. Read only what you're interested in. I also plan to frequently update this guide.

Tip: Zoom out if you can, the guide will be much more pleasant to read (ctrl + mouse wheel on desktop)

> So, first of all, who should read on, who am I addressing with this guide?

  • Those who are looking for a better roleplay experience, but haven't yet tried another LLM besides the default setting

  • Those who enjoy interacting with my bots, as I'm only testing them with DeepSeek (Scroll down to the bottom and read section IV. first)

When DeepSeek first came out, janitorai's reddit thread had started blowing up with posts about it. Everyone was raving about how amazing it is, people were encouraging each other to try it out, and many have created tutorials, tips and advanced prompts for it

Table of Contents:

I. General info on DeepSeek -> (What it is, why it is good)

II. Tutorial + Addressing issues -> (Skip here if all you want to know is how to set up DeepSeek)

III. Other links -> (may or may not be relevant to you, bunch of guides, fixes and advanced prompts)

IV. My own bots -> (Read if you like interacting with my bots)

Extra: Bot Scenario contains general comparisons between JLLM and DeepSeek. Example dialogs contain a bunch of advanced prompts: a general one, plus other prompts that caught my attention

Note: I create fandom-based bots here on janitor. Fallout (main), Game of Thrones, Squid Game, The Witcher, and more. If you're interested, you can check out my creator page here.

┈┈ • ୨ 🤍 ୧ • ┈┈

・・

✦•······················•✦•······················•✦

I. General Info on DeepSeek

・・

So what is DeepSeek?

DeepSeek is a company that develops multiple LLMs different to what janitorai.com uses by default

LLM stands for Large Language Model. In a nutshell: the LLM is responsible for forming the bot's reply. Different LLMs work and act differently. Janitorai.com allows its users to switch between different LLMs quite easily. This means that you can chat with janitor's bots using a different LLM, while remaining on the site, retaining the same interface and continuing your existing chat (although the difference will shine through much more if you start a brand new chat). Some people like using Claude or OpenAI LLMs but these are costly models. DeepSeek-R1 dropped in January 2025 and has quickly become quite popular. Since then, other models have appeared as well, and more will likely appear in the future, sometimes replacing the old models.

Like many others before me, I also find that DeepSeek is currently much better than janitorai's default LLM (namely, JanitorLLM or JLLM)

Let's see what makes it better (optional read):

If you open the Scenario of this bot, you will find Advantages/Disadvantages of DeepSeek that I gathered so far, based on posts on reddit, as well as my own experience. It also contains solutions to potential problems. Keep in mind that this section was written based on an older model, DeepSeek-R1.

┈┈ • ୨ 🤍 ୧ • ┈┈

・・

✦•······················•✦•······················•✦

II. Tutorial + Addressing issues

There are four steps to this:

1. Choosing a provider

2. Setting up DeepSeek

3. Adjusting Generation Settings

4. Getting an advanced prompt going (optional but recommended for an even better experience)

Find information about potential issues at the bottom of this section

・・

  1. Choosing a provider:

You can use DeepSeek through OpenRouter, Chutes, or directly through the Official DeepSeek API. Let's compare them so you can pick the one you prefer:

・・

OpenRouter characteristics:

  • free option: 50 messages / day / account (Note: This is the only truly free option! Also, abusing account creation gets you banned!) Better stick to a single account.

  • base tier: adding a one-time deposit of 10$ to your account balance allows you 1000 messages per day for a lifetime (rerolls are counted as 1 full message; make sure to use the "free" models — payed models will start deducting from your balance)

  • the money on your account may expire after a year but only if you don't use the service

  • I think you can withdraw your money any time as long as it hasn't expired

・・

DeepSeek official API characteristics:

  • no free trial

  • deposit any amount of money (minimum of 2$) and pay exactly per usage: see pricing details here. If it's confusing, pricing is explained here. (cost depends entirely on how heavily you use janitor, but I saw different people claiming that this version costs them ~0,5$ to 3$ per month)

  • the money on your account doesn't expire, but continues to be used up at the pace you're using the service

  • You cannot withdraw the deposited money

  • If you choose this option, you will automatically be using the FULL DeepSeek models, which give you better quality replies than the "free" models!

・・

Chutes characteristics:

  • no free trial

  • with the recent changes, this became the most expensive option

  • Prior to August 2025, Chutes required a one-time deposit of 5$ for 200 messages per day. Allegedly, those who made this payment in the past will still get access to their 200 daily messages without needing a subscription.

  • Current base subscription: deposit of 3$ / month allows you 300 messages per day (rerolls are counted as only 1/10 of a message; make sure to use the "free" models — payed models will start deducting from your balance), higher tiers start from 10$ / month

  • allegedly not the safest website (info leaks?) and they changed their pricing twice in the past 2 months, so caution is advised

・・

Once you've decided which you wanna use, follow one of the setup tutorials below!

❗️ Note: It is theoretically possible for providers to increase their prices in the future. Sadly, this is just how the world works, and I cannot garantee that it won't happen.

・・

  1. Setting up DeepSeek:

❗️ REMINDER: After you're finished setting up DeepSeek, make sure to come back here and check out step 3 and step 4 as well!

・・

OpenRouter setup:

・・

Official DeepSeek API setup:

・・

Chutes setup:

・・

❗️ IMPORTANT! — Save your API key! No matter what provider you use. Currently janitor saves the API key alongside your API settings, but still... you never know when you might need it.

Once you save your API settings, you may have to refresh the chat page before things start working

・・

  1. Adjusting generation settings (Goal is fluid, engaging, but controlled responses)

Go to the chat interface -> API Settings -> Generation settings

  • temperature: recommendation varies per model and provider

    • lower temperature is supposed to be more logical and stay better in character, while raising temperature is supposed to increase creativity (and chaos)

    • recommended temperature may vary — you gotta experiment or look around for suggestions (1.7-2.0 is what I usually use with the official API)

  • tokens: 0 (0 means unlimited and it helps avoid unfinished messages)

  • context size: 16k-32k (smaller context size might result in better roleplay quality)

Reference link here. You can read much more info on how to tailor your generation settings to fit your preference

・・

  1. Getting an advanced prompt (Optional but highly recommended to further enhance your experience)

Using an advanced prompt — especially a carefully customized one — can significantly improve the immersion and response quality

  • First, here you can read a general explanation on what advanced prompts and/or jailbreaks are (Though the example prompt this guide contains was not specifically made for DeepSeek). If you already know about advanced prompts, there's no need to open this guide

  • Now, setting up an advanced prompt for DeepSeek:

    If you open the Example dialogs of this bot, the first thing you find there is an advanced prompt that was designed to maintain realism, depth, and consistency in character behavior. It was not specifically designed for DeepSeek either, but will work with it regardless. I tested it. If you don't want to dive into advanced prompts further right now, simply copy pasting this prompt should be an acceptable starting point, I think

    However, if you want to further customise your experience, check the links under DeepSeek-R1 prompts in the III. Other links section below. I personally use Cheese's Prompts, I found they work beautifully with DeepSeek

Your advanced prompt goes to:

Chat interface -> Settings -> API settings -> Proxy configuration -> Custom prompt

・・

Congratulations! Your DeepSeek is ready, you can finally start your roleplay! 🧡

☕ If DeepSeek has "changed your life" and you found my guide useful, feel free to invite me for a coffee / tip me through my ko-fi donation page here :)

・・

If you're having any issues with DeepSeek:

  • Open bot Scenario. Starting from 11. under Disadvantages, I describe a bunch of potential problems and potential solutions, maybe you'll find yours there too

  • Check out this guide. Lots more potential solutions, especially regarding bot behaviour

  • Check out the III. Other links section below. Maybe one of them will be useful

  • If all else fails, you can ask on reddit or even here in the comments. I'm not an expert, far from it — but maybe I, or someone else can help you

┈┈ • ୨ 🤍 ୧ • ┈┈

・・

✦•······················•✦•······················•✦

III. Other Links

This section is just links I collected that you might find useful regarding janitorai and LLMs

The Ultimate Proxy Guide (massive but to to-the point, touches on several subjects; posted by a janitor MOD in July, 2025)

Proxy organization template (created by CelestialFirestorm on reddit)

JanitorAI current website server status and current JLLM context limit

What are Tokens? What is Context? And what happens if you hit it?

Chat Transplant Method - Continuing Roleplay after hitting Context <- this was useful with JLLM but I don't know if it's still relevant with DeepSeek (or maybe even counterproductive?)

Chat Memory Template

Visual Guide to DeepSeek Users (via chutes.ai) <- Sadly, they removed the post but the comments might still be of some use maybe

How to use DeepSeek Tutorial (via openrouter)

Molek's guide: Lots of amazing tips, solutions and advanced prompts for DeepSeek

DeepSeek FAQ

DeepSeek fixes & tips

Other LLM models / preventing DeepSeek's reasoning

Advanced Prompts

What are advanced prompts and how to use them? + JLLM custom prompts

DeepSeek-R1 advanced prompts

Additionally, the Example dialogs of this bot also contain other advanced prompts that caught my attention. I gathered them there for myself to test out later, but you can try them out too, if you want

JLLM advanced prompts

GPT advanced prompts/jailbreaks

┈┈ • ୨ 🤍 ୧ • ┈┈

・・

✦•······················•✦•······················•✦

IV. My own bots

Janitorai runs on JLLM by default. DeepSeek is a big improvement in comparison, so I stopped using JLLM and no longer test my bots with it. That said, I still keep token count reasonable so most of my bots remain JLLM compatible. Just a heads-up: If you use my larger bots (2k+ permanent tokens) on JLLM, you may need to manage chat memory carefully in case you’re doing a long roleplay.

This guide explains everything about DeepSeek, including setup — just scroll back up to the Table of Contents and pick what you’re interested in.

If you’re unsure whether to try DeepSeek or want to know what makes it so good, check out my JLLM vs DeepSeek comparison in the bot Scenario.

And whichever LLM you end up using, have fun roleplaying! :)

┈┈ • ୨ 🤍 ୧ • ┈┈

proxy allowed

Published chats

0

comments

Leave a comment or feedback for the creator ❤️