Janitor AI Proxy Guide 2026: Free Options, Setup & Risks

A Janitor AI proxy is a connection setting. It sends your chat to an outside model, such as DeepSeek or a model on OpenRouter, instead of Janitor’s built-in JLLM. You add a model name, a chat-completions URL and an API key in Janitor’s API settings. It’s optional: JLLM is free, and janitor+ Router now offers outside models without any key.

Use a proxy if you want a specific model, your own billing, or more control over generation settings. Skip it if you just want to chat. The setup takes ten minutes, but it moves your chats and your spending to another company.

Research-based guide, checked October 25, 2026 against Janitor’s help centre and changelog and against DeepSeek’s and OpenRouter’s documentation. We did not run an authenticated end-to-end test for this page.

Key Takeaways

• A proxy is a custom API connection: model name + chat-completions URL + your API key
• Free routes exist (OpenRouter :free models), but they’re capped at 50 requests a day until you buy credits
• Your chats go to the proxy provider under its terms, not Janitor’s
• janitor+ Router is the no-key alternative, but it needs a subscription and bills per token
• Never paste your key into a community “free proxy” link

What is a Janitor AI proxy?

In Janitor’s settings, “proxy” means a custom connection to an OpenAI-compatible model API. Janitor’s own help centre lists “proxy setup” among its official user guides. Proxies are a supported feature, not a hack.

There are now three ways to get a reply on Janitor:

1. JLLM, Janitor’s built-in model. It’s free, and janitor+ improves it.
2. janitor+ Router. Outside models, billed per token from a Janitor wallet. It needs a subscription and works on web only.
3. A proxy or custom API. Outside models through your own provider account, key and bill.

Proxy, direct API or janitor+ Router: which should you use?

Route	Cost	Setup	Who handles your chats	Best for
JLLM (free)	$0	None	Janitor	Most people
janitor+ Router	$12.99/month subscription plus per-token usage	Pick a model in API settings	Janitor plus the third-party model provider	Outside models without managing a key
OpenRouter proxy	Free :free models with daily caps, or paid per token	Account, key, URL, model ID	OpenRouter plus the model’s provider	Trying many models, including free ones
DeepSeek direct API	Prepaid, per token	Account, top-up, key, URL, model ID	DeepSeek	The lowest cost for heavy use

If you’re still deciding whether to pay Janitor at all, our Janitor AI subscription guide covers what janitor+ and Router cost.

How do you set up a proxy in Janitor AI?

These steps follow Janitor’s official proxy quickstart, updated to match its July 19, 2026 redesign of the API settings page:

1. Create an account with your provider (for example, OpenRouter or DeepSeek) and generate an API key. Treat the key like a password.
2. Open a chat on janitorai.com and go to the chat’s API settings.
3. Choose your provider under General. Janitor’s redesign moved provider selection there. Custom prompts, prefills and forbidden words now sit under Instructions.
4. Add a configuration: a name, the model name, the proxy URL and your API key, with no extra spaces. Leave the custom prompt blank for your first test.
5. Use the built-in connection test. Janitor’s May 3 update lets saved configs test their connection in place, and also adds import, export and recovery.
6. Click Save. The redesigned page has an explicit Save button that only lights up when something has changed.
7. Reload if Janitor asks you to. A banner appears when the active proxy URL changes, then you can send a test message.

Watch for: Janitor’s quickstart says to copy the model name and URL exactly and not to click the proxy URL. Pasting an extra character or space is the most common reason a new setup fails.

Which URL and model name do you use?

Provider	Proxy URL	Model name
OpenRouter	https://openrouter.ai/api/v1/chat/completions	Copy a current model ID from OpenRouter. Free variants end in :free
DeepSeek (direct)	https://api.deepseek.com/v1/chat/completions	deepseek-flash or deepseek-v4-pro

Janitor’s 2025 quickstart example used a DeepSeek V3 free model ID on OpenRouter. Free model IDs change often, so copy a current one from OpenRouter’s model list rather than an old guide. For DeepSeek’s own API, our DeepSeek on Janitor AI guide covers the key, balance, model names and current rates step by step.

What does a free Janitor AI proxy really cost?

The free route is real, but capped. OpenRouter’s documented limits for free models are:

• Per minute: 20 requests.
• Per day, before buying credits: 50 requests if you’ve bought fewer than 10 credits in total.
• Per day, after buying credits: 1,000 requests once you’ve bought at least 10 credits.

Janitor’s quickstart describes the same deal: 50 messages a day, reset at midnight UTC, or 1,000 a day after a one-time $10 payment. Every swipe and regeneration counts as a request, so 50 a day goes quickly in a long roleplay.

Paid routes bill per token. As of October 25, 2026, DeepSeek’s official Flash rates per million tokens were:

• Uncached input: $0.15 off-peak, $0.30 peak
• Output: $0.60 off-peak, $1.20 peak

Peak hours are 01:00–04:00 and 06:00–10:00 UTC on weekdays. Long chats cost more per reply, because the whole context is sent with every request. Our token pricing explainer shows how that adds up, and the proxy cost calculator below turns your own message count into a monthly figure.

Watch for: “free proxy” links shared in Discord servers, comments and videos. Anyone running a proxy can read every message that passes through it, and some ask for your own API key. That’s how keys get stolen.

What is the best API for Janitor AI?

It depends on what you’re optimizing for, so here’s the decision rather than a ranking:

• Lowest cost for heavy use: DeepSeek’s own API, with prepaid per-token billing at the rates above.
• Most model choice, plus a free tier: OpenRouter.
• No key and no setup, but paid: janitor+ Router.
• Free and simplest: JLLM, no proxy at all.

Whichever you pick, keep the prepaid balance small. If a key leaks, a small balance limits the damage.

Is using a Janitor AI proxy safe?

A proxy is as safe as the company at the other end. Your messages, character definitions and persona go to that provider under its terms.

Janitor’s privacy policy says it doesn’t cover third-party services. So once you switch on a proxy, the provider’s own policy decides how your chats are stored and used.

Practical rules:

• Use official provider endpoints only. These are the URLs in the table above. Avoid community “reverse proxies” run by strangers.
• Give Janitor a dedicated key. Don’t reuse it elsewhere, and delete it at the provider if you stop using it.
• Keep identifying details out of chats, whatever route you use.
• Check the provider’s policy before sending anything personal.

For the bigger picture, see our guide to what companion apps can see about you.

How do you fix common Janitor AI proxy errors?

Most proxy failures come down to the key, the balance, the model ID or a rate limit. The codes below are the ones DeepSeek and OpenRouter document.

What you see	Likely cause	Fix
401 or “authentication fails”	Wrong or disabled key, or an extra space	Re-copy the key; create a new one if needed
402 or “insufficient balance/credits”	Prepaid balance is empty	Top up at the provider
404 on OpenRouter, or 400/422 on DeepSeek	Old or mistyped model ID, or a setting the provider rejects	Copy a current model ID; reset generation controls to defaults
429 or “rate limit”	Too many requests; OpenRouter free models allow 20 a minute and 50 a day	Wait, slow down, or buy credits
502 or 503 on OpenRouter	The chosen model or its provider is down or unavailable	Switch to another model
500 or 503 on DeepSeek	Server error or overload	Retry after a short wait; check DeepSeek’s status page
403 on OpenRouter	Permission, guardrail or moderation block	Check the model’s restrictions or try another model
Worked yesterday, fails after a URL change	Settings saved but the page not reloaded	Reload when Janitor shows the banner

If JLLM also fails without the proxy, the problem may be Janitor itself. Janitor posted about errors affecting proxy users on June 10, 2026. Our Is Janitor AI down? checklist shows how to tell the difference, and the Janitor AI review covers the platform more broadly.

Sources & references

All checked October 25, 2026:

• TL;DR Quickstart Proxy Instructions: Janitor help centre, updated August 1, 2025
• We redesigned the API settings page: Janitor, July 19, 2026
• Janitor changelog: proxy config editing (May 3, 2026) and the proxy-error notice (June 10, 2026)
• janitor+ Router help article: Router eligibility and billing
• OpenRouter quickstart: API base URL and chat-completions endpoint
• OpenRouter rate limits: free-model caps
• OpenRouter errors: error codes
• DeepSeek models and pricing and DeepSeek error codes
• Janitor privacy policy: last updated September 17, 2026