🌌 Sunless — Complete Janitor AI Setup Guide

The easiest and most powerful free Gemini proxy for Janitor AI. No coding knowledge. No credit card. No Google login. Just copy-paste one command and start RPing.

100% Free & Unlimited for Personal Use:

• UNLIMITED Requests Per Day — It's localhost, go absolutely wild.
• No Account Risk — Runs in anonymous guest mode, your Google account is never touched.
• Auto-Updates — Every time you start Sunless, it automatically pulls the latest version from GitHub. You literally never have to update manually.
• Beautiful Web UI — Open http://localhost:5000 in your browser for a glassmorphic control center with model switching, tunnel management, and a built-in chat playground.
• Phone/Tablet Access — Built-in 1-click ngrok tunnel so you can RP on your phone while Sunless runs on your PC.

Side Notes:
1. Gemini models are mainly SFW and RPG focused, BUT whether NSFW works for you is NOT up to me — it's region-based. For example, in South East Asia it can pretty much do anything, but it may not work for you due to local restrictions. Your mileage may vary. Best model for NSFW: gemini-3.1-pro-thinking (less restrictions, deepest quality).
2. You need to start the server each time you wanna RP — or just keep it running, it's super lightweight. After the first install, just type nephis in any terminal and it launches instantly.

🥇 gemini-3.8-flash-thinking — Recommended Default. Super fast with thinking reasoning. Best balance of speed + quality.
⚡ gemini-3.1-pro-thinking — NSFW + Best Overall Quality. Less restrictions, deepest reasoning, richest worldbuilding.

• gemini-3.1-pro — Deep reasoning, no thinking tags. Slowest but highest quality.
• gemini-3.8-flash — Ultra-fast prose, no thinking overhead.
• gemini-3.7-flash — Fast frontier inference.
• gemini-3.7-flash-thinking — Chain-of-thought for intricate scenarios.
• gemini-3.5-flash — Reliable fast dialogue.
• gemini-3.5-flash-lite — Sub-second responses for quick turns.

💡 Thinking models (-thinking suffix) show internal reasoning inside <think>...</think> tags before the actual RP response. Makes roleplay way more nuanced for complex scenarios.

1. Download & install Termux (Play Store version is OUTDATED):
CLICK HERE TO DOWNLOAD TERMUX

2. Open Termux, paste this ONE command and press Enter (long-press to paste):

pkg update && pkg install python git -y && git clone https://github.com/InsomniacZero/Sunless.git && cd Sunless && bash start.sh

3. It will start the server immediately!

• In JanitorAI — Set Proxy URL to: http://localhost:5000/v1/chat/completions
• API Key: anything (literally type anything, e.g. gemini-rp)
• Model: gemini-3.8-flash-thinking
• Context Size: 16384 or 32768
• Max Output Tokens: 0 (this means unrestricted — DO NOT change it)
• Stop anytime: Press Ctrl + C (press Ctrl button in Termux UI, then C on your keyboard)
• Restart anytime: Just type nephis and press Enter!

1. Open PowerShell: Press Win + X → Click "Terminal" or "PowerShell".

2. Paste this ONE command and press Enter:

git clone https://github.com/InsomniacZero/Sunless.git && cd Sunless && .\start.bat

If the above doesn't work (PowerShell 5 users), try this instead:

git clone https://github.com/InsomniacZero/Sunless.git; cd Sunless; .\start.bat

⚠️ Don't have Git? Download it first from git-scm.com, install it, then try again.

3. It will start the server immediately!

• In JanitorAI — Set Proxy URL to: http://localhost:5000/v1/chat/completions
• API Key: anything
• Model: gemini-3.8-flash-thinking
• Context Size: 16384 or 32768
• Max Output Tokens: 0 (unrestricted)
• Stop anytime: Press Ctrl + C
• Restart anytime: Just type nephis in any terminal or PowerShell!

1. Open terminal and paste this ONE command:

git clone https://github.com/InsomniacZero/Sunless.git && cd Sunless && ./start.sh

2. Same Janitor AI settings as above:

• Proxy URL: http://localhost:5000/v1/chat/completions
• API Key: anything
• Model: gemini-3.8-flash-thinking
• Max Output Tokens: 0
• Restart anytime: Just type nephis

By default, Sunless only works on the device it's running on. If you want to RP on your phone or tablet while Sunless runs on your PC, you need a free ngrok tunnel.
⚠️ iPhone users: The Janitor AI website does NOT support localhost on Safari/iOS, so this section won't work for iPhones. macOS works fine though — just use the Mac setup above.

Step 1 — Get a Free Ngrok Account (1 minute, one-time):
1. Go to dashboard.ngrok.com/signup — sign up with Google or GitHub.
2. After signing in, go to Your Authtoken.
3. Copy the authtoken (looks like 2abc123def456_xxxxxxxxx).

Step 2 — Set Your Token in Sunless (one-time):
1. In your terminal type: nephis --menu
2. Select [5] Cloud Tunnel Control.
3. Select [2] Set Ngrok Authtoken.
4. Paste your token → press Enter.

Step 3 — Start the Tunnel:
In the same menu, select [1] Start Public Tunnel. You'll get a URL like:
https://abcd-12-34-56.ngrok-free.app

Step 4 — Use It on Your Phone:
On your phone, open janitorai.com → any character → API Settings.
Set Reverse Proxy URL to your tunnel URL + the path:
https://abcd-12-34-56.ngrok-free.app/v1/chat/completions
Set everything else the same (API Key: anything, Model: gemini-3.8-flash-thinking, Max Tokens: 0).

⚠️ The ngrok URL changes every time you restart the tunnel. Just copy the new one and update your Janitor AI settings.
💡 Keep Sunless running on your PC while you play on your phone — if you close it, your phone stops working too.

• PROXY ERROR 404 / "error not found":
You probably forgot /chat/completions at the end. Your full URL must be:
http://localhost:5000/v1/chat/completions

• "Network error occurred" / "rate limited" / "failed to fetch":
— Refresh your JanitorAI page 1-2 times.
— Make sure you're using Chrome browser (other browsers sometimes block localhost).
— Make sure Sunless is actually running in your terminal.

• Termux Errors / "Unable to locate package":
Type: cd ~
Then type: termux-change-repo
Select ALL mirrors/regions, press OK. Then redo the Android install command from above.

• "Connection refused" on phone:
— Make sure you set up the ngrok tunnel (see Phone/Tablet section above).
— Make sure Sunless is still running on your PC.
— Make sure you copied the full URL including /v1/chat/completions.

• Response is blank / model not found:
— Double-check your model name for typos. Copy it exactly from the model table above.
— Make sure Max Output Tokens is set to 0 (not blank, not 1, literally zero).

• Other stuff:
— Make sure there are no spelling mistakes in your API URL and model name.
— If nothing works, try stopping Sunless (Ctrl + C), then restarting with nephis.

❓ FAQ

Do I need a Google account / API key?
Nope. Sunless runs completely anonymously in guest mode. No login, no cookies, no API keys from Google.

Can Google ban my account?
Your account is never connected. At absolute worst, you might get a temporary cooldown (a few minutes). Just wait and try again.

Can I run this directly on my Android phone?
Yes! Use the Termux setup above. Both the server AND Janitor AI run on the same phone.

What's the <think> stuff in responses?
Thinking models (-thinking suffix) show their internal reasoning inside <think>...</think> tags before the actual RP response. Makes roleplay way more nuanced and contextual. The actual RP text comes after the thinking block.

Do I need to update manually?
No! Sunless auto-updates every time you start it. It pulls the latest version from GitHub automatically. You literally never have to run git pull yourself.

What is nephis?
After the first install, nephis is a shortcut command that launches Sunless from anywhere. Just open a terminal and type nephis — that's it.

Can you use this in lorebary ?

Yes. But the locahost link alone will NOT work, you need the ngrok tunnel. That's it, i dont use lorebary myself lol, you may ask in discord, because i dont use lorebary so idk much, others can help you there

✌️ Pre-Made Custom Prompt

Don't know what to put in your custom prompt / bot definition? I made one you can just copy-paste straight in:

CLICK HERE — Pre-Made Custom Prompt (Pastebin)

💡 Pro Tip: Enable Web Search Mid-RP

Whenever you send any message to Janitor AI, you can tell the model to use web search by just adding this at the end of your message:

OOC: Use web search

Like this:

Any problems or questions? DM me on Discord!