What Is a Proxy on Janitor AI? How It Works

A proxy on Janitor AI is a connection that lets the site use an outside language model, such as DeepSeek, instead of its built-in one. It is not a VPN and it does not hide your traffic. Janitor AI's own glossary calls it "the middleman between your chat and the model provider": you enter a provider's web address and your API key, and Janitor AI sends your chats there. This guide explains what that means, how the pieces fit, what it costs and how to stay safe. Everything is checked against Janitor AI's help centre and the providers' price pages on October 7, 2026.

What does "proxy" mean on Janitor AI?

On Janitor AI, a proxy is a setting that points the site at another company's AI model. Janitor AI is a character chat platform. By default it answers with JLLM, the Janitor Large Language Model, which its help centre describes as "the default model used in Janitor unless you've hooked up an external one via API". A proxy setting replaces that default with a model you choose, such as DeepSeek, GPT or Claude.

The word causes confusion because "proxy" normally means something else. In networking, a proxy server relays your traffic to hide or route it. On Janitor AI the word is shorthand for "an OpenAI-compatible API address". Nothing is hidden or rerouted. Your chat goes to the address you typed, the model replies, and Janitor AI displays the answer.

Janitor AI's glossary uses a kitchen comparison: if JLLM is the default kitchen, the proxy lets you order takeout from another restaurant.

How does a Janitor AI proxy work?

A Janitor AI proxy works in three steps: Janitor AI sends your chat to the web address you configured, the model provider generates a reply, and Janitor AI shows it. The three parts are the front end, the endpoint and the model.

Part	What it is	Example
Front end	Janitor AI, the chat interface you click on	janitorai.com
Endpoint	The web address that accepts chat requests	https://api.deepseek.com/chat/completions
Model	The AI that writes the reply	deepseek-flash

You give Janitor AI four values in its API settings, as the official quickstart shows: a configuration name, a model name, a proxy URL and an API key. The key is a secret password that tells the provider whom to bill. All three of the key, URL and model must come from the same provider. A DeepSeek key will not work against another company's address.

This works because most providers copy the request format that OpenAI popularised. DeepSeek's documentation says its API is compatible with OpenAI and Anthropic formats, which is why Janitor AI can talk to it by changing only the address and the model name.

Proxy, VPN and API: what is the difference?

A proxy on Janitor AI sends chats to a model, a VPN hides your internet address, and an API is the general door that makes the proxy possible. They solve different problems and do not replace each other.

Term	What it does	Needed for Janitor AI proxy?
Proxy (Janitor AI sense)	Sends chat requests to an outside model	This is the setting itself
API	A programming interface that lets one service call another	Yes, the provider exposes one
API key	A secret that identifies and bills your account	Yes
VPN	Hides your IP address by routing traffic elsewhere	No

Janitor AI's glossary defines a VPN as something that "hides your real IP address and reroutes your internet through a different location". You do not need one to use a proxy configuration.

Why use a proxy instead of the default model?

People add a proxy for three reasons: model choice, memory length and control. The default model is free and works out of the box, but it has a limited memory. Janitor AI's token guide says JLLM's memory holds between 8,000 and 9,000 tokens, after which it forgets the oldest parts of the chat.

	Default JLLM	Proxy to an outside model
Cost	Free	You pay the provider per token
Model choice	One built-in model	Any compatible model
Memory	About 8,000 to 9,000 tokens, per Janitor AI	Set by the model; DeepSeek supports a 1M-token window
Setup	None	API key, URL and model name
Who sees your chats	Janitor AI	Janitor AI and the model provider

The last row is easy to miss. With a proxy, your messages reach a second company. Read the provider's privacy terms before you send anything you would not want stored.

What does a proxy cost?

A proxy costs whatever the model provider charges per token, and the range between models is wide. You pay for every message, because each one resends the character definition and chat history as input, then bills the reply as output. The table assumes 8,000 input tokens and 300 output tokens per message with no caching. That is an assumption, because Janitor AI's real prompt size depends on its settings.

Model	Input per 1M tokens	Output per 1M tokens	Cost per 1,000 messages
OpenAI GPT-6 Luna	$0.10	$0.50	$0.95
DeepSeek deepseek-flash, off-peak	$0.15	$0.60	$1.38
DeepSeek deepseek-flash, peak	$0.30	$1.20	$2.76
Anthropic Claude Haiku 4.5	$1	$5	$9.50
Anthropic Claude Sonnet 5.5	$2	$10	$19.00
Anthropic Claude Opus 5.5	$4	$20	$38.00

Prices come from the DeepSeek, OpenAI and Anthropic price pages. Caching lowers the real DeepSeek figure a lot, because DeepSeek bills repeated prompt prefixes at a fraction of the normal rate. Our DeepSeek pricing breakdown explains the peak and off-peak hours and the cache rules in detail.

Whether a model works through Janitor AI depends on whether its provider accepts the OpenAI-style request format. DeepSeek does. For others, a router such as OpenRouter puts many models behind one OpenAI-style address, which is why the official quickstart uses it. If you are weighing routers, see whether to keep routing through OpenRouter after the Stripe acquisition.

For a broader look at why the cheapest listed price is not always the cheapest bill, read why the cheapest AI API is not the cheapest to run.

Is there an easier option than a proxy?

Yes, if you subscribe. Janitor AI's Router lets Janitor+ subscribers use supported third-party models without an external proxy or API keys, with per-token prices shown before you chat. Per the help page, last updated July 31, 2026, it is web only for now and is not a replacement for the built-in JLLM+. For the full picture of plans and models, see what Janitor AI is.

Is a Janitor AI proxy safe?

A proxy is safe when you use a trusted provider and protect your API key, and risky when you use an unknown "free proxy" site. Three risks matter.

1. Key theft. Anyone who gets your key can spend your balance. Never paste it into bot descriptions, comments or screenshots. Create a separate key for Janitor AI so you can delete it without breaking anything else.
2. Untrusted endpoints. A stranger's free proxy address sees every prompt and the key you send it. Use the provider's own address or a router you have checked.
3. Data handling. Your chats go to the provider. Providers differ on retention and training. Read whether the free AI API tier trains on your data before you choose.

Keep the balance on your account low, so a leaked key has a capped cost. Check usage on the provider's dashboard regularly.

Why does my proxy show an error?

Most proxy errors come from a wrong URL, a wrong model name, an empty balance or a missing page refresh. Janitor AI's error guide recommends the same first steps for every error: check the configuration says Active, click Save Settings, refresh the whole page, then click Test.

Message	What it usually means
"A network error occurred"	The URL is wrong, or you did not refresh after saving
401	The key is wrong for that provider
402	The provider balance is empty
404 "No endpoints found"	The model name is wrong or retired
429	Too many requests, or a daily cap was reached
500 or 503	The provider is overloaded; wait and retry

DeepSeek lists the same codes with their causes on its error page.

How do you set one up?

The exact addresses for DeepSeek are in DeepSeek proxy URL for Janitor AI.

For a full walkthrough with exact values and costs, follow our DeepSeek on Janitor AI setup guide.

Pick a provider, create an API key, then add a configuration in Janitor AI's API Settings with the model name, proxy URL and key. The exact values depend on the provider. For DeepSeek the model name is deepseek-flash and the URL is https://api.deepseek.com/chat/completions. Older guides that use deepseek-chat or a free OpenRouter model are out of date. Save the settings and refresh the page before you test.

Sources

Janitor AI glossary: What is JLLM, Janitor AI: Tokens guide, Janitor AI: Quickstart proxy instructions, Janitor AI: OpenRouter error guide, DeepSeek: Your First API Call, DeepSeek: Models and Pricing, DeepSeek: Error Codes, OpenAI API pricing, Anthropic pricing.