Janitor AI Not Working: Find the Broken Link First
Work out which part of the chain broke before you start changing settings. A sixty second test that tells you whether it is Janitor, your provider, or one wrong field.
Most guides hand you ten things to try in order and hope one of them lands. That wastes an evening. Janitor AI is a chain of four parts, and once you know which one snapped, the fix is usually a single field. Here is how to find out in about a minute.
The sixty second test
Before changing anything, switch the chat over to the built in JanitorLLM and send one message.
- JanitorLLM works, your own model does not. The site is fine, your internet is fine. The problem is in your API configuration, and nothing you do to your browser will help.
- Both fail. The problem is Janitor AI itself, your connection, or something local like an extension or a firewall.
- Both work, but only on some characters. The card is the problem, not the platform. Usually an oversized definition or a broken prompt on the creator's end.
That one test splits the entire problem space in half and takes less time than reading the rest of this page. If it points at Janitor rather than your setup, our Janitor AI status page will confirm whether others are seeing the same thing.
What each symptom actually means
| What you see | What it usually is | Fix |
|---|---|---|
| Failed to fetch, network error, load failed | Janitor cannot reach the endpoint you gave it | Check the base URL and model ID against your provider's current documentation |
| Worked earlier, stopped mid session | Provider credit ran out, or the conversation outgrew the context window | Check your balance first, then start a fresh chat |
| 401 | Key, URL and model name are not all from the same provider | Re-enter all three from one place |
| 402 | Billing. No credit behind the key | Top up. It is not a connection problem despite how it reads |
| 429 | Rate limited, daily or per minute | Covered in detail on the 429 page |
| 500 or 502 | Provider or platform side failure | Nothing to fix from your end. Wait, or switch model |
| Replies arrive with visible thinking text | You are on a reasoning model | Switch to a standard chat model |
| Blank reply, no error | Content filter on the provider side, or a malformed card | Try the same message on a different character |
The five things that cause most of it
Mixing providers
The single most common configuration error. People take a base URL from one provider, a key from another, and a model name from a guide written about a third. All three have to come from the same place. When they do not, the failure often surfaces as a generic network error rather than anything that names the real cause.
Running dry mid session
If it worked for twenty messages and then stopped, that is almost never a technical fault. Pay as you go credit does not warn you before it runs out, and the failure arrives as an error rather than a notice.
Retired model names
Model identifiers get deprecated. An ID copied from a guide written six months ago may simply not exist any more, and providers rarely fail gracefully when that happens. If your setup worked for months and broke without you touching it, check whether the model you are calling still exists under that name.
The conversation outgrowing its window
Every model has a context limit, and a long roleplay eventually exceeds it. Symptoms range from outright errors to the character quietly forgetting things it knew an hour ago. If failures started appearing deep into a long chat and a fresh chat works, this is why.
Local interference
Worth ruling out only after the sixty second test points at the platform rather than your API settings. Ad blockers and privacy extensions can intercept the calls, antivirus and Windows Firewall occasionally block the site outright, and a badly out of date browser fails in ways that look like server problems. A VPN can cause it or cure it, so try both states rather than assuming either.
What is worth your time and what is not
Clearing your cache, switching browsers and toggling a VPN are the standard advice everywhere, and they do occasionally work. They are also the reason people lose whole evenings, because they get tried first against problems they cannot possibly fix. If the sixty second test says your API configuration is wrong, no amount of browser fiddling will help. Diagnose, then act.
Or remove the chain entirely
Every failure on this page exists because Janitor AI asks you to run the model plumbing yourself. That is real control, and for some people it is the entire point. For everyone else it is a maintenance job attached to a hobby.
RolePlaya runs the models server side. No key, no base URL, no billing account, no deprecated model IDs, and nothing to reconfigure when a provider changes something. The free tier has a daily message allowance and no advertising.
If you want the full comparison rather than the sales pitch, including what you lose by giving up model choice, it is written up here.
Common questions
Why does Janitor AI say network error?
Usually because it could not reach the endpoint you configured. A wrong base URL, a wrong model name, or a key with no credit behind it all surface this way. Switching to the built in model tells you immediately whether the problem is your configuration or the platform.
Why does it work at first and then stop?
Two likely causes. Your provider credit ran out mid session, which fails without warning, or the conversation grew past the model context limit. Check your balance first, then try a fresh chat.
Is Janitor AI down or is it just me?
Switch to the built in model. If that works, the platform is up and the problem is your setup. If nothing works at all, check a status page before spending time on your own configuration.
Why are my replies blank?
Either a content filter on the provider side rejected the generation, or the character card itself is malformed. Sending the same message to a different character separates the two in one step.
Will clearing my cache fix it?
Only if the fault is genuinely local, which is the minority of cases. If the built in model works while your own does not, the problem is in your API settings and no amount of browser maintenance will change that.
Janitor AI, JanitorLLM, OpenRouter, DeepSeek, KoboldAI and all other names used on this page are trademarks of their respective owners. RolePlaya is not affiliated with, endorsed by, sponsored by or connected to any of them. Platform behaviour and error handling described here reflect publicly documented information as of September 2026 and change without notice.
