The Absolute Beginner's Guide to Using a Proxy with Janitor Part Two

Psst...you didn't skip Part One of this guide...did you?

Proxy Guide, Part 2: Leveling Up Your AI Experience

Welcome back! By now, you’ve successfully set up your first proxy using OpenRouter and have seen how different AI models can transform your chats. You've graduated from the basics.

Now, let's level up. In this part, we'll explore:

• Going Paid: Unlocking the best quality and unlimited messages.
• The Art of Prompting: How to control the AI and get the responses you want.
• Deeper Troubleshooting: Solving new problems as you explore more powerful models.

Section 1: Going Paid - Why and How

The free tiers are fantastic, but you might want more—more messages, higher quality, or access to the absolute best models. Paying for a proxy service is the way to do it, and thankfully, it can be incredibly cheap.

Why Pay?

1. Quality: Paid models are often smarter, more creative, and better at staying in character.
2. Stability: Free services can get overloaded. Paid services are faster and more reliable.
3. Freedom: No more counting messages. Chat as much as you want!

There are two main ways to go paid. Let's look at them.

Option A: The Official DeepSeek API (The Best Quality)

This is the cheapest and highest-quality way to use DeepSeek models. You are getting your AI directly from the source.

How does pricing work?

It's like a pre-paid phone card. You "top up" your account with a few dollars, and your balance is used up very, very slowly as you chat. A $5 top-up can last for thousands of long messages.

Setup Guide: DeepSeek Official API

Follow this guide to get yourself setup.

Stage 1: Getting Your DeepSeek Key

1. Go to DeepSeek: Open a new tab and go to https://platform.deepseek.com/
2. Create an Account: Sign up for a free account.
3. Top Up Your Account: Navigate to the "Usage" tab on the left. You'll need to add a balance. The minimum is usually $2, but adding $5 is a great start.
4. Create Your Key: Go to the "API Keys" tab. Click "+ Create new key", name it something like "JanitorAI," and click "Create".
5. COPY YOUR KEY! Just like before, copy this new key (it will start with sk-...) and save it somewhere secure, ike a digital notepad app.

Stage 2: Setting up JanitorAI

1. Open API Settings: Back in JanitorAI, go to a chat and open the API Settings.
2. Select Proxy: and select "+ Add Configuration" on the top right, if you already have openrouter setup from Part One.
3. Enter a Configuration Name: This could be "Offical Deepseek".
4. Enter the Model Name: This would be the version of DeepSeek you've chosen, like "R1-0528".
5. Enter the API URL: This is the address of the DeepSeek service. In the "Proxy URL" box, paste this exact URL: https://api.deepseek.com/v1
6. Paste Your Key: In the "API Key" box, paste your new DeepSeek key that you just saved.
7. Choose the Model: You have to type the model name manually. These are the two best options: For the smart, reasoning model (R1): deepseek-reasonerFor the creative, softer model (V3): deepseek-chat
8. Save and Refresh! Hit "Save Settings" and refresh the page.

You are now connected directly to DeepSeek! Enjoy the premium quality.

Option B: Using Paid Models on OpenRouter (The Most Variety)

If you like the convenience of OpenRouter but want to try paid models, it's easy. This is how you can access premium models like Anthropic's Claude.

1. Add Credits: Go to your OpenRouter account and add some credits to your balance.
2. Change the Model Name: In your Janitor.AI API settings (using your same OpenRouter key), simply choose a paid model. The key difference is that paid models do not have :free at the end of their name.

• Example: deepseek/deepseek-chat-v3-0324 (This is the paid version)

Trying the Luxury Models: Claude

OpenRouter also gives you access to Claude models by Anthropic, which are famous for their beautiful writing style. They are more expensive, so keep an eye on your credits, but they are fantastic for rich, narrative-driven stories.

• Try anthropic/claude-3.7-sonnet for a great balance of quality and cost.

Model Deep Dive

DeepSeek Family

V3-0324: Your reliable friend

• Agreeable, moldable to your style
• Good for slice-of-life, romance, lighter content
• Sometimes too accommodating

R1-0528: The intense one

• Shows reasoning in <think> tags
• Follows character traits to the extreme
• Better for dramatic, action-heavy scenarios
• Can be "meaner" if character calls for it

R1T-Chimera: The experimental mix

• Combination of V3 and R1 traits
• More unpredictable results
• Only available on some proxies

Gemini Models

2.5-Pro: The scholar

• Incredible knowledge base
• Excellent at author mimicry
• More passive, wants you to lead
• Best for fandom characters, historical figures

2.5-Flash: The quick student

• Faster responses
• Good enough for most uses
• Can get repetitive in long chats

Claude Models

Sonnet: The novelist

• Beautiful, literary writing style
• Excellent character consistency
• More expensive but worth it for quality
• Good at slow-burn romance

Opus 3: The master storyteller

• Exceptional creative writing ability
• Deep understanding of narrative structure
• Excellent at complex character development
• Superior reasoning and problem-solving
• Very expensive but unmatched quality
• Best for: Premium roleplay experiences, complex plots, nuanced character interactions

Haiku: The economical choice

• Faster and cheaper than other Claude models
• Decent quality but not exceptional
• Can feel somewhat dry or mechanical
• Best for: Testing setups, quick interactions, budget-conscious users

Understanding Free vs Paid Models

What You Get for Free

• OpenRouter: 50 messages/day (1000 if you add $10 once)
• Google Gemini: 100-1000 messages/day (depending on model)
• Mistral: Nearly unlimited, but lower quality

Why Consider Paying?

• Reliability: Free services can go down or change limits
• Quality: Paid models often write better and remember more
• Speed: Less waiting, fewer "server busy" errors
• Features: Access to the newest, most advanced models

Alternative Proxy Services

Google Gemini (Free Trial)

Best for: Users who want high-quality free access

Setup Overview:

1. Create Google Cloud account
2. Enable AI Studio API
3. Get free $300 credit (lasts 3 months)
4. Configure JanitorAI with Gemini models

Pros: Excellent writing, great memory, smart reasoning

Cons: Requires Google account, has content filters

Chutes (Paid)

Best for: Heavy users who reroll messages frequently

Unique features:

• Rerolls count as 1/10th of a message
• Good for specialized DeepSeek variants
• One-time $5 payment for daily allowance

Section 2: The Art of Prompting

You have a powerful AI. Now, how do you tell it what to do? With a prompt!

A prompt is a set of instructions you give the AI that it reads before every single reply. It's the best way to shape the bot's personality, writing style, and behavior.

You place your prompt in Janitor.AI's "Custom Prompt" box in the API Settings.

Your First "Starter" Prompt (Copy and Paste this!)

Here is a great, all-purpose prompt to get you started.

[System Note: Your primary goal is to portray {{char}} with depth and consistency. Write in a descriptive, narrative style from a third-person limited perspective. Focus on {{char}}'s internal thoughts, feelings, and reactions and avoid writing for or controling {{user}}'s actions, thoughts, or dialogue. Avoid clichés and repetitive phrases. Stay in character at all times, adhering strictly to the personality defined in your character sheet.]

Breaking it Down:

• [System Note: ...]: This tells the AI these are important, high-level instructions.
• portray {{char}} with depth: Encourages the AI to go beyond surface-level personality traits.
• narrative style from a third-person limited perspective: This defines the writing style, like a novel.
• avoid writing for or controling {{user}}: This prevents the AI from deciding what your character does or says—a very common and annoying problem.
• Avoid clichés: Helps keep the writing fresh.

Feel free to experiment! Change words, add rules (like "Use asterisks for actions"), or remove things you don't like. This is your main tool for controlling the AI.

For a much deeper dive into prompting, check out Advanced Prompting 101!

Section 3: Deeper Troubleshooting

You've explored new models and now you're seeing new behaviors. Here’s how to fix them.

Problem: My bot is showing <think>...</think> tags!

• What it is: You're using a "reasoning" model like deepseek-reasoner or deepseek/deepseek-r1-0528. The AI is showing you its thought process before generating the reply.
• The Fix: You can simply ignore it! It doesn't affect the roleplay. If it bothers you, you can edit it out of the bot's message. A good prompt can sometimes reduce it, but it's a natural feature of these models.

Problem: The bot is getting repetitive or boring.

• The Fix 1: Edit the last message. Delete the repetitive part of the bot's last reply and write a more interesting sentence to guide it in a new direction.
• The Fix 2: Increase Temperature. In the Generation Settings, nudge the Temperature setting up by 0.1. This encourages the AI to be more creative and less predictable. Don't go too high (above 1.4) or it might become nonsensical.
• The Fix 3: Be more descriptive yourself! The AI mirrors your energy. If you give detailed, engaging replies, it will too.

Problem: The bot is forgetting important details.

• The Fix 1: Check Context Size. As mentioned in Part 1, make sure your context size is set to 16,384 for the best balance of memory and performance.
• The Fix 2: Use Chat Memory. At the top of the chat, next to the three dots, is the Chat Memory feature. Use it to write down key plot points, facts, or relationship details (e.g., "Character A and B just had their first kiss."). The AI reads this before every message. It's like a cheat sheet for your story!

Advanced Configuration Tips

Context Management

Sweet Spot: 16,384 tokens
Why: Best balance of memory vs performance
Avoid: 32k+ (models get confused with too much info)

Temperature Settings by Model

DeepSeek V3: 0.8-1.0 (let it be creative)
DeepSeek R1: 0.6-0.8 (keep reasoning focused)
Gemini: 0.7-0.9 (balanced creativity)
Claude: 0.6-0.8 (maintains writing quality)

For more information about temperature settings and it's effects, check out Understanding Temperature!

Custom Prompts

Now that you're comfortable with basics, consider adding custom prompts:

For more vivid writing:

Write with rich sensory details and emotional depth. 
Show don't tell. Use varied sentence structure.

For character consistency:

Stay true to the character's established personality, 
speech patterns, and background throughout the conversation.

Cost Management Strategies

Budget-Friendly Approach

1. Start with free tiers to test preferences
2. Mix free and paid: Use free for casual chats, paid for important scenes
3. Monitor usage: Check your spend regularly
4. Use discount hours: DeepSeek is 75% off during certain hours

Smart Spending Tips

• Shorter contexts = lower costs
• Avoid excessive rerolling with paid models
• Use free models for testing character setups
• Save paid models for your favorite characters

Understanding Model Personalities

When to Use Which Model

DeepSeek V3 - Choose when you want:

• Cooperative, agreeable interactions
• Slice-of-life scenarios
• Characters that adapt to your energy
• Romantic, fluffy content

DeepSeek R1 - Choose when you want:

• Characters with strong, defined personalities
• Dramatic or intense scenarios
• Action scenes with consequences
• Characters that challenge you

Gemini - Choose when you want:

• Accurate fandom portrayals
• Educational or historical content
• Complex world-building
• Characters with deep knowledge

Claude - Choose when you want:

• Beautiful, literary prose
• Emotional depth and nuance
• Slow-burn character development
• Premium writing quality

Optimization Checklist

So, after all this reading, and practicing, and playing around, you can now:

• Switch between different models easily
• Understand your daily/monthly limits
• Recognize which model fits different scenarios
• Adjust temperature and context appropriately
• Monitor your usage and costs (if paying)

You're no longer a beginner! You understand the landscape and can make informed choices.

🎯 Quick Action: Try switching between DeepSeek V3 and R1 with the same character. Notice how different they feel? That's the power of understanding your tools!

Updated on: 02/08/2025