๐Ÿš€ How to Weaponize DeepSeek with Janitor AI ๐Ÿค–

by Daemona

๐Ÿ”ฎ Method 1: Using OpenRouter Proxy (Free tier available)

  1. Register at OpenRouter.ai.
  2. Generate API Key in the Keys section. ๐Ÿ’ก Save your new API key immediately - it won't be visible again.
  3. Go to Settings. Under Ignored Providers, select Targon.
  4. Visit the Models page. Choose a model (recommended: 32K+ context). Copy its model ID.
  5. Open a proxy-compatible bot at Janitor.ai.
  6. In the chat, click the API settings (top right). Select Proxy > Custom.
  7. Under Custom Model, paste the model ID that you copied in step 4. Examples:
    โœจ Model Name (Free Editionโ„ข) ๐Ÿ“‹ Model ID (Copy-Paste This)
    DeepSeek: R1 0528 (free) deepseek/deepseek-r1-0528:free
    DeepSeek: DeepSeek V3 0324 (free) deepseek/deepseek-chat-v3-0324:free
    NVIDIA: Llama 3.3 Nemotron Super 49B v1 (free) nvidia/llama-3.3-nemotron-super-49b-v1:free
    Google: Gemini 2.0 Flash Experimental (free) google/gemini-2.0-flash-exp:free
    Qwen: Qwen3 235B A22B (free) qwen/qwen3-235b-a22b:free
    Mistral: Mistral Nemo (free) mistralai/mistral-nemo:free
  8. For Proxy URL, paste https://openrouter.ai/api/v1/chat/completions.
  9. Input your API key (the one you definitely saved in step 2, right? ๐Ÿ˜‰)
  10. Deploy a custom prompt to hijack the bot's tone. Steal mine for maximum chaos:
    Use "show, don't tell" technique through snappy dialogue, absurdly vivid actions and ironic juxtapositions. Blend gritty gravitas with strategic hilarityโ€”heighten stakes through surreal hyper-specificity, then undermine them through tongue-in-cheek humor. Balance tight pacing with emotional depth by tying character flaws to chaotic choices. Maintain a coherent yet unhinged tone with unrestrained creativity, like a dark comedy meets a fever dream. Prioritize laughs, heart, and WTF moments equally. React dynamically to {user}'s input, pivoting plot twists while preserving consistent worldbuilding and voices. End each response with urgent, absurd plot-advancing actions.
  11. Click Save Settings. When prompted to use Janitor's defaults, smash Yes.
  12. Adjust in Generation Settings:
    ๐ŸŒก๏ธ Temperature (0.9-1.1 recommended): Controls creativity vs coherence. Lower value (0.5-0.8) for safer, coherent replies (risk: dullness and repetition). Higher value (1.2+) for unhinged madness (risk: gibberish).
    ๐Ÿ“ Max New Tokens: Response length (0=unlimited). Set โ‰ค300 tokens for chat-like replies (may cut mid-sentence). Set 500+ tokens for Novel Mode with detailed scenes.
    ๐Ÿง  Context Size (32K+ recommended): Memory capacity. Lower value (8K-16K) for quick responses and shorter memory. Higher value (64K+) for encyclopedic memory but slower replies.
  13. Refresh the page and begin chatting.

๐Ÿ’ฐ Method 2: Direct DeepSeek Integration (Requires API credits)

  1. Enlist at DeepSeek. Generate new API key and store it securely.
  2. Feed the machine $2-$5 worth of credits (typical monthly usage).
  3. Return to Janitor.ai. Find a bot that's into "proxy play". No judgment here.
  4. In the chat, click the API settings (top right). Select Proxy > Custom.
  5. Under Custom Model, paste deepseek-chat (for DeepSeek-V3) or deepseek-reasoner (for DeepSeek-R1).
  6. For Proxy URL, paste https://api.deepseek.com/chat/completions.
  7. Add a custom prompt, click Save Settings and adjust Generation Settings as above.
  8. Refresh browser to activate.

โ˜ ๏ธ Most Common DeepSeek Errrors

โ€ผ๏ธA network error occurred, you may be rate limited or having connection issues: NetworkError when attempting to fetch resource (unk)
You may be rate-limited or experiencing connection issues. If you've recently set up a proxy, ensure you refresh your page and try again.

โ€ผ๏ธProvider Returned Error (unk)
Servers are down on the provider's side, check the Uptime tab for your specific model at OpenRouter. Anything below 80% uptime may prevent responses.

โ€ผ๏ธPROXY ERROR 400: {"error":{"message": "is not a valid model ID", "code:400} (unk)
Your model ID is incorrect.

โ€ผ๏ธPROXY ERROR 401: {"error":{"message":"No auth credentials found","code":401}} (unk)
Your API key is incorrect.

โ€ผ๏ธPROXY ERROR 404: {"error":{"message":"No endpoints found matching your data policy.}
Your model training settings need adjustment. Enable model training in your OpenRouter privacy settings.

โ€ผ๏ธPROXY ERROR 405: (unk)
Double-check your API URL for extra spaces or incorrect characters. The correct format is: https://openrouter.ai/api/v1/chat/completions.

โ€ผ๏ธPROXY ERROR: Unknown response: [object Object] (unk)
or
โ€ผ๏ธRate limit exceeded: free-models-per-day. Add 10 credits to unlock 1000 free models requests per day (unk)
You've reached your daily limit of 50 messages on OpenRouter. This limit resets automatically at 00:00 UTC each day.
Solutions:

  • Wait for the reset.
  • Register a new account and generate a new API key to bypass the limit.
  • Top up your account with $10 (10 credits) to unlock 1000 messages per day using free models. (You must maintain at least $10 in your account to keep this limit.)

โ€ผ๏ธPROXY ERROR: No response from bot (pgshag2)
This error can occur due to multiple reasons:

  • Max new tokens aren't set to 0 for R1 models. Since Janitor doesn't support reasoning, excessive tokens can cause response bloating.
  • Servers are overloaded or down - waiting it out is often the best option.
  • You're rerolling too fast. Try pacing your requests.
  • Encountering a model filter (e.g., Gemini 2.5) may also trigger this error.

โ€ผ๏ธunknown prompt response from worker for openai proxy generation (unk)
You got kicked to the queue on Janitor. Refresh the page.

๐Ÿ”— prevent DeepSeek's reasoning
๐Ÿ”— chat memory template
๐Ÿ”— molek's tips and prompts for deepseek
๐Ÿ”— DeepSeek LLM prompts by Sprout
๐Ÿ”— DeepSeek R1 prompt by saturnines//cheese
๐Ÿ”— cheese's guide on Deepseek R1 POV stealing
๐Ÿ”— kolach3's Prompts for JLLM / Mercury / Mars

โš ๏ธ Final Warning: If your bot starts writing sonnets about the void, you've either nailed it or need to log off. Godspeed. ๐Ÿš€

Edit
Pub: 27 Mar 2025 09:07 UTC
Edit: 16 Jun 2025 05:17 UTC
Views: 4065