JanitorAI JLLM or a proxy? How to decide which to use

Two ways to get a reply

Every JanitorAI chat needs something to write the bot's side. JLLM is the default answer: it is Janitor's own model and works the moment you open a bot. A proxy is the other answer: you point Janitor at an outside provider, and that provider's model writes the replies instead.

Your bots, chats and persona stay on JanitorAI either way. The choice only decides who writes the next message, so it is cheap to try one, then the other, and come back.

Side by side

The proxy fields themselves (URL, key, model) are explained in what a proxy is in JanitorAI, so this page sticks to the decision.

Five questions that settle it

• Do you care which model writes your story? If "whatever works" is the honest answer, JLLM saves you the setup.
• Do you want control over replies beyond the bot itself? A proxy pairs with a custom prompt and your own generation settings, so you can shape length, point of view and pacing for every chat.
• Are you comfortable keeping an API key safe? A key is like a password for your provider account. If that sounds like a chore, start on JLLM.
• Whose privacy terms do you want to read? With a proxy, each message also goes to that provider, so their policy matters as much as Janitor's.
• Will you stick with long stories? Long chats are where model choice and settings like context size show most, so heavy players tend to care more.

Test both on one chat

Opinions about models are cheap; your own bot is the real test. Pick a chat you know well, ideally one with a tense or emotional moment near the end.

• On JLLM, swipe the last reply three times and note what you like: voice, length, whether the bot stays in character, whether it writes your actions for you.
• Switch the provider to your proxy, save, reload the page and swipe the same reply three times again.
• Compare the six replies side by side. Three swipes each is enough to see a pattern without fooling yourself with one lucky reply.

If you can't decide, keep the proxy configuration saved and flip between them per bot. Some people keep JLLM for quick, light chats and a proxy for the long story they care about.

If you pick a proxy

You need a provider, a key and a model name. With UnsavedAI a free account gets you a key in the Your keys tab after you make one at Setup, and gemma-4-12b works on every plan; Priority adds gemma-4-31b. We never store API traffic and nobody reads or screens it. Every field is shown in the JanitorAI connection guide, and the numbers to start with are in JanitorAI generation settings.

Looking at other free options too? Read free proxies for JanitorAI before pasting a key anywhere.

Questions people ask