JanitorAI slow or not responding? Why and what to try
Updated
First work out which kind of slow you have. If pages load but a reply crawls, it is the model: JLLM or your proxy's provider. If the reply never arrives, it is stuck: copy your message, reload and send once. If pages themselves won't load, it is the site. With a proxy, lower context size, keep Max new tokens moderate and stop stacking swipes. On UnsavedAI, Priority skips the queue.
Try it now, free, no account neededStart a story with Empress SuzuranAn empress stopped time. You just walked in.Slow, stuck or down?
- Slow: the reply arrives, just late. Something is queueing or reading a big prompt.
- Stuck: the typing indicator runs on and nothing ever comes, or the reply stops halfway and never finishes. The request died quietly.
- Down: the JanitorAI pages themselves will not load or log you in. That is the site, and what to do when JanitorAI is down covers it.
Most "JanitorAI is slow" moments are the first kind, and the cause depends on who writes your replies.
If you use JLLM
JLLM is JanitorAI's own built-in model, so its speed is out of your hands. Your options are to wait and try again in a few minutes, or switch the chat to a proxy so a provider you choose writes the replies. JLLM or a proxy walks through that choice.
If you use a proxy
With a proxy, Janitor just passes your message on and waits for the provider, so the usual causes sit there: - The provider is busy and your request is waiting in its queue. - The chat is long, so each request carries tens of thousands of tokens the model must read before it writes a word. Why the first reply after a break is slowest is in slow first replies over an API. - Max new tokens is set very high and the model is writing a very long reply. - Your own requests are stacking: swipes fired while a reply is still coming, or the same chat open in two tabs. - Your connection is weak, especially on a phone switching between Wi-Fi and mobile data.
Unstick a reply in four steps
- Give it a minute or two. Long replies in long chats can genuinely take that long.
- Copy your last message so it is not lost.
- Reload the page and reopen the chat.
- Send once, or swipe once if the stuck reply was the bot's.
If a red error appears instead of a reply, it names the problem: 5xx proxy errors, 429 too many requests or Failed to fetch.
Settings that make replies quicker
| Setting | Change | Why it helps |
|---|---|---|
| Context size | Lower it if you don't need deep history, and keep facts in chat memory | Less text to read before writing starts |
| Max new tokens | 400 to 600 instead of 1,000 or more | Shorter replies finish sooner |
| Streaming, if your settings offer it | On | Words appear as they are written, so you are not staring at a blank box |
| Swipes and tabs | One at a time, one tab | Your own requests stop queueing behind each other |
Lowering context means older messages fall out of view sooner, so pair it with a good chat memory entry.
Waiting on a UnsavedAI key
On UnsavedAI, Free writes one reply at a time and Priority two, and at busy times Free requests wait behind Priority ones. If speed is what you are missing, Priority skips that queue for $4.99 a month; if waiting runs long you get a message saying so rather than endless typing. Models shows whether gemma-4-12b and gemma-4-31b are online right now.
Questions people ask
Why is JanitorAI slow only at night?
Evenings are when most people chat, so models and providers are busiest. The same chat is often much quicker earlier in the day.
Does a long chat make JanitorAI replies slower?
With a proxy, yes: the whole context is sent with every message, so a very long chat means more to read before the reply starts.
Should I keep swiping when a reply is slow?
No. Each swipe is another request, and they queue behind the slow one. Let it finish or fail first.
