Janitor AI Proxy Setup 2026: OpenRouter, DeepSeek, Gemini, Fixes

To set up a proxy on Janitor AI, create an OpenRouter API key. Then, in a Janitor chat, click the purple "using janitor" button in the header, set Provider to Proxy, click "+ New" and enter a name, the proxy URL https://openrouter.ai/api/v1/chat/completions, your key and a model ID you copied from OpenRouter's live model list that day. Finally click Add and refresh the page.

Updated October 28, 2026. The model IDs and free limits below were checked against OpenRouter's public model list and docs on that date, and the Janitor screens were checked in Janitor's live interface the same day.

This Janitor AI proxy guide covers the setup end to end, how DeepSeek and Gemini fit in, what the free limits really are, and what each common error means. If you want Grok specifically, read the Janitor AI Grok API guide. If Janitor itself will not load, start with Janitor AI not working.

What a Janitor AI proxy is (and what it is not)

On Janitor AI, "proxy" means a model provider you choose writes the replies instead of Janitor's built-in model. Janitor still holds your characters and chats. The provider receives the prompt for each reply and, if the model is paid, bills your account for it.

OpenRouter is the most common choice because one key reaches models from many companies, including DeepSeek and Google. It is not a free unlimited AI, and it is not part of Janitor: it is a separate service with its own account, limits and privacy settings. Janitor's changelog also lists other providers that work as a proxy, such as Chutes and NanoGPT (added June 2026). The steps are similar, but the URL and model names differ.

What you need before you start

Step 1: Create an OpenRouter API key

Sign in to OpenRouter, open the API keys page and create a new key just for Janitor. Give it a clear name. If you plan to add credits, set a credit limit on the key so one runaway chat cannot spend your whole balance. Copy the key once and keep it private. Never paste it into a public doc, a Discord message or a "shared proxy" someone else runs.

Step 2: Pick a model ID that exists today

This is where most setups break. Janitor's own quick-start article still shows the example deepseek/deepseek-chat-v3-0324:free. On October 28, 2026 that ID was not in OpenRouter's model list, so copying it leads to a "No endpoints found" error.

Open the models page on OpenRouter, search for what you want and copy the ID exactly, including the part before the slash. What the list showed on October 28, 2026:

Free model IDs end in :free, and the free list changes often. Treat the table as an example, not a promise, and copy from the live list.

Step 3: Open Janitor's settings and fill in the proxy form

Open a chat and click the purple using janitor button in the chat header. A settings panel opens on the side with three tabs: Settings, Memory and Customize. The proxy options are on the Settings tab.

The panel shows two notices worth reading first: after you add or change a proxy URL, you must refresh the page for it to take effect, and Janitor has no affiliation with any proxies, so judge the provider yourself.

First, set Provider to Proxy. The dropdown lists five options: janitor, janitor+ Router, Proxy, OpenAI and Claude.

The Provider dropdown in Janitor's chat settings panel. Screenshot of Janitor's interface taken on October 28, 2026; the interface may change.

Then, under "Proxy configurations", click + New. With no saved setup, the panel says "No configurations yet" (add one or import JSON), and "+ New" opens a "New Configuration" form with these fields:

The empty "New Configuration" form. Screenshot of Janitor's interface taken on October 28, 2026; the interface may change.

The URL must be exactly that, with no extra slash and no repeated segment. If you typed only https://openrouter.ai/api/v1, the "Add /chat/completions" button next to Proxy URL appends the rest of the path. A Show button sits next to API key; use it only when nobody can see your screen. Janitor's error guide names a wrong URL as the most likely cause of a network error.

Step 4: Test, add and refresh

While the form is open, the panel says "Save or cancel proxy config first". Once Name, Proxy URL, API key and Model are filled in, the Test button becomes available; use it before you add the configuration. Then click Add. As Janitor's note asks, refresh the whole Janitor page, reopen the chat and send one short message to a simple character. The panel also has its own Save button at the bottom for its other settings; if your Provider choice does not stick after the refresh, save the panel and check again. Add your custom prompt, memory and long cards only after a plain message works, so you know which change broke things.

Using DeepSeek on Janitor AI

DeepSeek works through the same four steps; only the model ID changes. As of October 28, 2026 every DeepSeek model on OpenRouter is paid, so you need credits on your OpenRouter account first. Guides that promise a free Janitor AI DeepSeek proxy are usually showing an ID that has since been removed. Pick a current ID, such as a flash model for faster, cheaper replies or a larger model for long scenes, and compare them on the same ten turns before you settle.

Using Gemini on Janitor AI

Gemini models are also on OpenRouter, and they were also paid on October 28, 2026. Do not confuse Gemma with Gemini: Google's Gemma models have :free versions, but they are smaller open models, not Gemini. Janitor's error guide notes that models like Gemini can fail silently when a prompt trips their filters, so if Gemini returns nothing in one scene, test a neutral message before you blame the setup.

Free model limits on OpenRouter

If you stay on :free models, OpenRouter's limits page listed these caps at the time of writing:

Janitor's guide adds two details: failed requests count toward the daily cap, and the counter resets at 00:00 UTC. In a long roleplay, 50 requests can go in one evening, because every reply, retry and swipe is a request.

Fix proxy errors: error, cause, fix

Match the message you see, then apply the fix. Quoted messages come from Janitor's OpenRouter error guide; the meaning of 401 comes from OpenRouter's error docs.

When the reply is empty or cut off

An empty reply with no error is a different problem. OpenRouter's docs say a request can succeed but return no text, for example while a model is starting up, or when a reasoning model spends all of its max tokens on thinking before it writes. Raise the max tokens setting if Janitor's settings panel shows one, or switch to a non-reasoning model. If replies stop mid-sentence, the same max tokens setting is usually the cause.

Keep your key and your chats safe

A proxy adds two companies to your chat: OpenRouter and the model provider behind it. The prompt for every reply, including the character card and recent messages, goes to them. Read OpenRouter's privacy settings before you change them to fix a 404, set a credit limit on your key, and delete the key when you stop using it. Avoid public "free proxy" links run by strangers: you cannot see who logs your chats or your key.

Proxy setup or a hosted chat like LumiChat

A proxy makes sense if you enjoy choosing models and tuning prompts. If you mainly want to talk to characters without keys, credits or error codes, a hosted service is simpler. LumiChat runs the model for you in the browser, and if you have a SillyTavern character card (JSON or PNG) that you made or are allowed to use, you can import it into LumiChat. LumiChat does not import chat histories from other sites. Browse the LumiChat character library or compare options in the Janitor AI alternatives guide. If you stay on Janitor, the Advanced Prompts guide and the Chat Memory guide help once your proxy works.

FAQ

What is the Janitor AI proxy URL for OpenRouter?

Use https://openrouter.ai/api/v1/chat/completions exactly, with no extra slash.

Is there a free DeepSeek proxy for Janitor AI?

Not on OpenRouter as of October 28, 2026: every DeepSeek model there is paid. Free models end in :free and change often.

Can I use Gemini on Janitor AI for free?

Gemini models on OpenRouter were paid on October 28, 2026. Gemma :free models are free, but they are not Gemini.

How many free messages do I get?

OpenRouter's free models allow 20 requests a minute and 50 a day, or 1000 a day once you have bought at least 10 credits. Failed requests count too.

Why do I get error 429 on Janitor AI?

Either you hit the free daily cap or the provider behind the model is busy. The message text tells you which one.

Why does Janitor AI say "No endpoints found"?

The model ID you entered is gone or offline. Copy a current ID from OpenRouter's model list.

Is it safe to use a Janitor AI proxy?

It is as safe as the services your chats go to. Use your own key with a credit limit, avoid public proxies, and remember that OpenRouter and the model provider receive your prompts.