A Janitor AI proxy is a connection that lets your Janitor chats run on an outside AI model, such as DeepSeek or Claude, instead of Janitor’s free built-in model. You set it up in API Settings with three things: a model name, an API URL and your own API key from a provider. This guide shows the setup with OpenRouter and the official DeepSeek API, what it costs, who sees your chats, and how to fix the errors people hit most.
What is a Janitor AI proxy?
Janitor is the front end: the chat window, the characters and the settings. The model is the part that writes the replies. By default that is JanitorLLM (JLLM), Janitor’s own model, which is free and built in. Janitor’s help center describes a proxy as “the middleman between your chat and the model provider”: you point Janitor at another company’s model and pay that company, or use its free allowance, directly.
Every reply the model writes is built from more than your last message. Janitor sends the character’s personality and scenario, your persona, the chat memory, any custom prompt and as much recent chat as fits. That matters for both cost and privacy, as you’ll see below.
| Option | What it is | Cost | Who processes your messages |
|---|---|---|---|
| JLLM | Janitor’s own model, the default | Free; janitor+ ($12.99/month plus tax) adds 5x context | Janitor |
| janitor+ Router | Third-party models inside Janitor, no API key needed (web only) | janitor+ plus per-token charges from an in-app wallet | Janitor and the model provider |
| Your own proxy | Your API key from a provider, pasted into API Settings | Free tiers with daily caps, or pay per token | Janitor, the proxy service and the model provider |
People switch for longer memory, a different writing style, or a model that follows instructions better. Janitor’s own Router help page still recommends JLLM+ for everyday roleplay, so a proxy is an option, not an upgrade everyone needs.
Before you start: Janitor’s guides are out of date
Janitor’s quickstart guide (last updated August 2025) still tells you to use the model name deepseek/deepseek-chat-v3-0324:free on OpenRouter. At the time of writing, OpenRouter’s model API lists no available endpoints for that model, so copying it exactly gets you a “No endpoints found” error. The free DeepSeek names on its DeepSeek troubleshooting page, such as deepseek/deepseek-r1:free, have no endpoints either, and OpenRouter lists no free DeepSeek model at all. Its older part-two guide, since removed from the help center, used deepseek-chat and deepseek-reasoner on the official DeepSeek API; DeepSeek’s changelog set July 24, 2026 as the date those names would be discontinued, and its current model list no longer includes them.
The steps below follow Janitor’s guides for where things go, with model names current as of September 2026. Model lists change every few weeks, so always copy the name from the provider’s own model page.
How to set up a proxy on Janitor AI with OpenRouter
OpenRouter is the service Janitor’s own guides use. One key gives you access to hundreds of models from different providers, including some free ones.
Step 1: Get an OpenRouter API key
- Go to openrouter.ai and create a free account.
- Open your profile menu, then Keys, then Create Key. Name it something like “Janitor”.
- Set a credit limit on the key if you plan to add money. It caps what that key can ever spend.
- Copy the key (it starts with
sk-or-v1-) and store it somewhere safe. You only see the full key once.
Step 2: Add the configuration in Janitor
- Open a chat with any character on janitorai.com.
- Open the menu at the top of the chat and choose API Settings.
- Choose Proxy, then + Add Configuration.
- Fill in the fields as shown below.
- Make sure the configuration shows as Active, click Save Settings, then refresh the whole page. Janitor’s guides call the refresh “crucial”.
- Send a test message.
| Field | What to enter |
|---|---|
| Config Name | Anything, e.g. OpenRouter DeepSeek |
| Model Name | The exact model ID from OpenRouter’s models page, e.g. deepseek/deepseek-v4.1-flash (paid) or openrouter/free (a free router) |
| Proxy URL | https://openrouter.ai/api/v1/chat/completions |
| API Key | Your sk-or-v1- key, with no spaces before or after |
| Custom Prompt | Leave blank at first (see below) |
Free models end in :free. openrouter/free picks one of the currently free models at random, so the writing style can change from reply to reply. For a consistent voice, pick one specific model and stick with it.
Setting up the official DeepSeek API
Janitor’s older part-two guide also covered going direct to DeepSeek. You pay DeepSeek per token from a prepaid balance, with no middle service.
- Create an account at platform.deepseek.com and top up a small balance.
- Under API Keys, create a key (it starts with
sk-) and copy it. - In Janitor, add a new proxy configuration. Model Name:
deepseek-flash(DeepSeek V4.1 Flash) ordeepseek-v4-pro. Proxy URL:https://api.deepseek.com/chat/completions, the endpoint in DeepSeek’s docs. Janitor’s older guide shows a/v1/version of the same address. - Paste the key, save, and refresh.
DeepSeek’s docs say both models support a thinking mode. Reasoning output can show up as a block of the model’s working before the reply; Janitor’s guide says you can ignore it or edit it out.
Context size, custom prompt and temperature
Context size is how much text the model reads each turn: character, prompt, memory and recent chat. JLLM’s free context is roughly 8,000 to 9,000 tokens (about 1,000 tokens per 750 words). Models reached through a proxy can take far more; DeepSeek’s current models list a 1M-token context. Janitor’s beginner proxy guides recommended about 16,384 tokens for most models and warned that 32k and up can make them slower and more confused. On a paid model, context size is also your biggest cost lever, because every token you send is billed every turn.
Custom Prompt is the field in your proxy configuration for a system prompt: instructions the model reads before every reply. Use it for style and roles, not for trying to switch off a provider’s rules. A simple starting point:
You write {{char}} and the world around them in third person, present tense. {{user}} is played by the human, who writes all of {{user}}'s words, actions and feelings. Keep replies to 1 to 3 paragraphs and end on {{char}}'s action, line or question.Our Janitor AI prompts guide has more copy-paste prompts for pacing, length and chat memory.
Temperature controls randomness. Janitor’s older guide suggested nudging it up by 0.1 if replies get repetitive and not going above 1.4. On DeepSeek’s own API, temperature has no effect while thinking mode is on (its default).
What a Janitor AI proxy costs
- OpenRouter free models: 50 free-model requests a day, or 1,000 a day once you’ve bought at least 10 credits (US$10), and 20 requests a minute. Per Janitor’s error guide, error replies count toward the daily cap, which resets at midnight UTC.
- OpenRouter paid models: the provider’s per-token price with no markup, plus a 5.5% fee ($0.80 minimum) when you buy credits by card.
- DeepSeek direct: per-token prices with an off-peak rate at half the peak rate. Peak is 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays.
Here is what one reply costs at a 16,000-token context and a 400-token reply, using list prices as of September 2026:
| Model | Price per 1M tokens (input / output) | Per reply | Per 100 replies |
|---|---|---|---|
DeepSeek deepseek-flash, direct, peak | $0.30 / $1.20 | about $0.005 | about $0.53 |
DeepSeek deepseek-v4-pro, direct, peak | $1.32 / $3.96 | about $0.023 | about $2.27 |
| Claude Sonnet 5 via OpenRouter | $2.00 / $10.00 | about $0.036 | about $3.60 |
DeepSeek’s off-peak rates halve its rows, and its cheaper cached-input rate can lower them further. On DeepSeek direct, thinking mode is on by default, and the reasoning it writes is billed as output on top of the visible reply. Rerolls cost the same as new replies. Doubling the context roughly doubles the input cost.
For comparison, janitor+ costs $12.99 a month plus tax and includes a Router balance; Router then bills per token at the rate shown for each model. For what stays free on Janitor itself, see is Janitor AI free.
Privacy: who sees your chats on a proxy
With a proxy, every message in that chat, plus the character, persona and memory Janitor sends with it, goes to the proxy service and the model provider behind it. Each has its own logging, retention and training policy, on top of Janitor’s.
- Janitor: its privacy policy lists chat conversations and messages as data it collects and stores.
- OpenRouter: its docs say it doesn’t store your prompts or responses unless you opt in, though it keeps request metadata such as token counts. It passes each request to a model provider, and each provider has its own policies. Your privacy settings control whether it may route to providers that may train on your prompts, with separate settings for free and paid models.
- DeepSeek: per its privacy policy, it processes and stores personal data in the People’s Republic of China, and lists training and improving its models among the uses of data it collects.
One trade-off to know: Janitor’s troubleshooting guides tell you to turn on OpenRouter’s model training setting to fix some free-model errors. That fix works because it allows routing to providers that may train on what you send. Decide whether that’s acceptable before you flip it. Our is Janitor AI safe guide covers Janitor’s own data handling in detail.
Why to skip “free” reverse proxies and shared keys
Ask around for a free Janitor AI proxy and you’ll be pointed to reverse proxies: a stranger’s server, shared in a comment or a Discord, that forwards your requests to a paid model. Some also pass around API keys that were never theirs to share. Both are a bad deal.
- The operator sees everything. Every message, your persona and the character’s full definition pass through their server, with no privacy policy and no one accountable.
- They may be running on someone else’s bill. “Free” access to a paid model can mean a leaked or stolen key. Using it means running up charges on another person’s account.
- They disappear. When the key is revoked or the owner shuts it down, your setup stops working mid-story.
- Your own key is at risk too. DeepSeek’s API terms say you’re solely responsible for fees caused by sharing or leaking your key. Never paste your key into a third-party site, “key checker” or shared proxy.
The legitimate free options are JLLM itself and the free models on a named service like OpenRouter. If you think a key has leaked, revoke it on the provider’s dashboard and create a new one. Deleting your Janitor account doesn’t revoke it.
Troubleshooting common Janitor AI proxy errors
Janitor’s error guide says to run this checklist first: make sure the configuration shows Active, click Save Settings even if nothing changed, refresh the whole page, then reopen the configuration and click Test. If you still get an error, match it below.
| Error | Usual cause | Fix |
|---|---|---|
| 400 Bad Request, or “maximum context length” | Typo in the model name or URL, or more tokens than the model accepts | Re-copy the model ID and URL; lower Context Size or trim the character and memory |
| 401 Invalid credentials | Wrong, incomplete or disabled key | Paste the full key with no spaces, or create a new one |
| 402 Payment required | Out of credits, or a negative balance | Add credits, or switch to a free model |
| 403 Forbidden | The provider’s moderation or guardrails blocked the request, or the key lacks permission | That model’s usage policy doesn’t allow the request; pick a model whose policy fits your story |
| 404 “No endpoints found” | The model ID no longer exists | Copy a current ID from the provider’s models page |
| 404 “No endpoints found matching your data policy” | Your OpenRouter privacy settings exclude every provider for that model | Pick a model with a provider that meets your settings, or change them knowing the trade-off above |
| 408 or 502 | Timeout, or the model is down | Wait and retry, or switch models |
| 429 “free-models-per-day” | Daily free cap reached | Wait for the midnight UTC reset, or buy credits |
| 429 “rate-limited upstream” | The free model is overloaded | Try another model or a paid version |
| “A network error occurred” | Wrong URL, or no refresh after saving | Check the URL exactly, save, refresh |
If the bot forgets things or repeats itself on a proxy, check Context Size first, then use Chat Memory for key plot points.
Skip proxies altogether with Cherrypop
API keys, token math and error codes are a lot of setup for a chat. Cherrypop.ai is the easiest way to get uncensored AI chat and roleplay with none of it: no keys, no model IDs, no bills from a third-party provider. You get a big catalog of ready-made characters (female, male and trans, realistic and anime, across many personality types), a create-your-own wizard, and AI images of your character, NSFW included. It’s free to start, and paid plans unlock more messages, images and features (see pricing on the site). It runs in the browser, including on mobile, and it’s 18+ only, with age checks where the law requires them. For more options, see our list of Janitor AI alternatives.
FAQ
What does proxy mean on Janitor AI?
It means using an outside AI model through your own API key instead of Janitor’s built-in JLLM. The proxy is the connection between your Janitor chat and that model’s provider.
Which proxy should I use for Janitor AI?
OpenRouter is the easiest start, since one key covers free and paid models and it’s the service Janitor’s own guides use. Going direct to a provider such as DeepSeek cuts out the middle service. janitor+ subscribers can use Router with no key at all.
How do I turn on a proxy on Janitor AI?
Open a chat, go to API Settings, choose Proxy and add a configuration with a model name, proxy URL and API key. Make sure it shows as Active, save, and refresh the page. To go back to the free model, switch the API setting back to JLLM.
Is there a free proxy for Janitor AI?
Yes, within limits. OpenRouter’s free models allow 50 requests a day, or 1,000 after a one-time purchase of $10 in credits (plus OpenRouter’s purchase fee). Avoid shared reverse proxies and leaked keys for the reasons above.
Does Janitor AI watch your chats?
Janitor’s privacy policy says it collects and stores your chats, and it doesn’t say whether staff can view private ones. With a proxy, the proxy service and model provider receive your messages too, under their own policies.
Is using a proxy on Janitor AI allowed?
Yes. Janitor documents proxy setup in its own help center. Each provider’s usage policy still applies to what you send through it.