Deepseek proxy setup

This is not an actual roleplay bot. (you can still try, lol)

Here is a step by step guide on how to setup Deepseek proxy without chutes or openrouter.

Step 1: Get your DeepSeek API Key

1. Go to the DeepSeek Platform. (https://platform.deepseek.com)

2. Top up your account (usually $2–$5 is plenty for millions of words).

   • Note: DeepSeek's "Free" tier is often heavily rate-limited, so a tiny paid balance makes it much smoother.

3. Go to API Keys and create a new one. Copy it immediately.

Step 2: Configure Janitor AI

1. Open your API Settings in Janitor AI.

2. Set "API Setup" to OpenAI (even though it's DeepSeek, we use this because the formats match).

3. In the Proxy URL field, paste:

   https://api.deepseek.com/chat/completions

4. In the API Key field, paste your DeepSeek key.

In the Model field, type:

deepseek-chat (This is for the V3 model)

Or use deepseek-reasoner for the R1 "thinking" model.

If it looks like that you're golden.

Step 3: Save and Refresh

• Click Save Settings.

• Crucial: Refresh your browser tab.

Step 3: Generation settings

• Go to Generation Settings.

• Set your Max Output Tokens a bit higher (around 800–1000) to give it breathing room.

• Context Size (16,384): This is very safe. It’s plenty of "memory" for the bot to remember what happened a few dozen messages ago. Since DeepSeek is cheap, you can actually slide this up to 32k later if you find the bot is "forgetting" the beginning of a long story.

• For deepseek-reasoner, it’s best to keep your Temperature between 0.6 and 0.8. If you set it too high (like 1.5+), the "reasoning" part of the brain gets a bit fried and the bot might start looping or talking nonsense.

• If you're using the reasoner, the bot might take 5–10 seconds before it starts typing. Don't worry—it’s not frozen! It’s just calculating the logic and character consistency behind the scenes. This usually leads to way fewer plot holes in roleplay.

The above is what I use, once it is working and generating chat you can play around with it the hard part of settings is done. Context and tokens will cost money but temperature will not.

You're paying directly to Deepseek they have a low rate of just parking a minimum $2 instead of $10 like Chutes.