Published November 20, 2026
Janitor AI Proxies Explained _ What They Are and How to Set One Up
"Proxy" is the most confusing word in Janitor AI. It is not a VPN, it does not hide your IP, and you do not need one to use the site. A Janitor proxy is simply a custom API connection: a setting that sends your chats to an outside model provider instead of Janitor's built-in JanitorLLM. This page explains what that means, when it is worth it, and how to set one up without breaking your chats.
Why proxies exist at all
Janitor AI is a front end. The characters, the interface, and your account live on Janitor. The actual replies have to come from a language model somewhere. JanitorLLM is the house option: free, unlimited, no key, no card. A proxy is the bring-your-own option. You open an account with a provider, get an API key, and tell Janitor to send your chats there. You pay that provider per message, and in return you get its models: bigger context windows, different writing styles, sometimes stronger memory.
There is also a middle route now. janitor+ Router, part of the paid janitor+ tier at $12.99 a month, offers outside models billed from a Janitor wallet, with no key of your own to manage. If you want better models without touching API settings, that is the route built for you.
The three pieces of a proxy
Every proxy setup is the same three fields, and all three must belong to the same provider:
- Proxy URL — the provider's chat-completions address. For OpenRouter it is https://openrouter.ai/api/v1/chat/completions.
- Model name — the exact model ID, copied from the provider's model list. One wrong character and nothing works.
- API key — your private key for that provider. OpenRouter keys start with sk-or-v1-.
Mixing pieces across providers is the number one setup failure. A DeepSeek key with an OpenRouter URL will never connect. When something breaks later, these fields are also where most API errors come from, so learn to recognise them now.
Setup, step by step
- Create your provider account, add a little credit if it has no free tier, and generate an API key.
- Open any chat on janitorai.com and open the API settings from the top-right menu.
- Choose the proxy or custom provider option and add a configuration with a name you will recognise.
- Paste the model name, proxy URL, and API key exactly, with no extra spaces.
- Leave the custom prompt blank for your first test. Save. Reload if the site shows a banner asking you to.
- Send a short test message. If a reply lands, you are connected.
One practical tip: save each provider as its own named configuration instead of overwriting one setup. When a model breaks mid-chat, you can swap to your spare configuration in seconds instead of retyping URLs and keys from memory.
OpenRouter is the easiest starting point because one key opens many models, including free variants whose IDs end in :free. Know the cap before you rely on it: free models on OpenRouter are limited, commonly 50 requests a day, and error messages can count against you. DeepSeek direct is the cheap single-model route many roleplayers settle on. The full field-by-field version of this setup lives in the homepage walkthrough.
The risks nobody mentions
A proxy moves your chats and your spending to another company. That provider sees your messages and applies its own privacy policy and content rules, which may be stricter than Janitor's. Your key is a password for your wallet: anyone who has it can spend your credit. Never paste it into a site other than your provider and Janitor's own settings, and never share it in a screenshot.
Be most careful with "free proxy" URLs posted in comments, Discord servers, and random guides. A stranger's proxy can log everything you write and harvest keys pasted alongside it. If a proxy costs nothing and asks for your key, you are the product. Stick to providers you can name, with real documentation and a billing page.
So, do you need one?
For most people, no. JanitorLLM is free and, with sane settings, good enough for everyday roleplay. A proxy earns its keep when you hit JLLM's real limits: the roughly 9,000-token memory, or a writing style you cannot get out of it. Start free, learn the settings, and add a proxy only when you can name the exact problem it solves for you.