JANITOR AI FREE PROXY (CLOUDFLARE)

openrouter acting up? yeah same.
this setup is better for most of you anyway.

this uses:

• google gemini (flash / lite)

• cloudflare worker (as proxy)

• works directly in janitor like openai

⚡ WHY THIS OVER OPENROUTER?

• hundreds–thousands of messages/day free (make extra keys under new projects for more)

• faster streaming (no delay mid-response)

• handles long chats + lorebooks WAY better

• more stable for RP bots

🧩 STEP 1: GET GEMINI API KEY

go here:
https://aistudio.google.com/app/apikey

create a key and copy it.

don’t lose it. if you do, that’s on you.

WANT MORE MESSAGES?
limits are per project, not per key.
same gmail → create a new project in AI Studio → make a new key under that project.
each project gets its own free quota.
you can make several projects (usually up to ~10).

🧩 STEP 2: GET THE PROXY CODE

copy this:
https://pastebin.com/LJc8e8mP

(go inside this link)

you don’t need to understand the code. just copy the whole thing.

🧩 STEP 3: CLOUDFLARE WORKER SETUP
(the most common place people mess up)

go here:
https://dash.cloudflare.com/

then follow exactly:

1. create a free account (or log in)

2. on the left menu click Workers & Pages

3. click Create application → Create Worker

4. give it any name (example: my-gemini-proxy)

5. click Deploy (this creates the empty worker)

6. after it deploys, click Edit code

7. delete ALL the default code (select everything and delete)

8. paste the full code you copied from pastebin

9. click Deploy / Save and Deploy

after deploying, you’ll get a link like:

https://your-name.workers.dev

this is your proxy URL.
copy it carefully. no spaces at the end.

🧩 STEP 4: CONNECT TO JANITOR AI

go to:
Settings → API Settings

set this:

API Type:
Proxy

API URL:
https://your-name.workers.dev (the exact link you got)
(DO NOT add /chat/completions or anything else)

API Key:
(your gemini api key – paste it clean, no extra spaces)

🧩 STEP 5: MODEL

use one of these (copy exactly):

gemini-3.5-flash
or
gemini-3.5-flash-lite

(if one is busy / rate limited, try the other)

Other models you can try : gemini-3-flash-preview , gemini-3.1-flash-lite
, no extra spaces)

🧩 STEP 6: CUSTOM PROMPT

For peak Roleplay experience

https://pastebin.com/QK8dQ27G

Paste the prompt you get from this link in the Global Prompt section of API settings and youre good to go !

⚡ RECOMMENDED SETTINGS

• temperature: 1

• max tokens: 0 (infinite)

• streaming: ON

• Do NOT use advanced settings

💀 COMMON ERRORS & FIXES

PROXY ERROR: No response from bot (pgshag2)

• most common error. means janitor got nothing back from the proxy.

• check: is the worker URL correct? (no extra /chat/completions, no spaces)

• did you actually click Deploy after pasting the code?

• wrong or expired API key → make a new key

• gemini rate limited you → wait a bit or switch to another key/project

• sometimes just high traffic → wait 1–2 minutes and try again

Proxy Error / Connection Error / Failed to fetch

• worker URL is wrong or has a typo

• you forgot to deploy the worker

• cloudflare worker is down (rare) → re-deploy it

• your internet / VPN is blocking it → try without VPN

Model not found

• you typed the model name wrong

• use exactly: gemini-3.5-flash or gemini-3.1-flash-lite-preview

• no extra spaces, no “models/” in front

[GOOGLE API BLOCKED IT]

means gemini didn’t like the prompt (safety filter)

fix:

• rephrase slightly

• don’t go full unhinged on the first message

• try regenerating or switching model

Traffic / Rate limit / Resource exhausted

• you hit the free daily/minute limit on that project

• fix: wait (resets around midnight Pacific) or make a new project + new key

• don’t share one worker/key with many people – it kills the quota for everyone

⚠️ IMPORTANT

DON'T use someone else’s worker

seriously.

if 100 people use one:

• it slows down

• breaks

• stops working

make your own. it takes 2 minutes.

🧠 FINAL

this setup is:

✔ free (hundreds–thousands msgs/day, with extra projects)
✔ fast
✔ great for heavy bots / lorebooks
✔ better than openrouter for most people

only downside:
gemini models can change names sometimes. if one stops working, just ask in comments for the current name.

but so does everything else.

that’s it. if you mess this up, reread step 3 carefully.