The architecture is what sets it apart. Janitor hosts the characters, the chat interface and a free built-in model called JanitorLLM. But power users connect an external model of their choice, and DeepSeek through OpenRouter is the community favorite, via an API key or a reverse proxy. The platform stays free. You pay the model provider for what you use, which for budget models works out to a few dollars lasting weeks.

That design buys enormous freedom and, honestly, some friction. Newcomers who expect an app-store experience bounce off the setup. People who push through get roleplay tunable to whatever model they can afford, with none of the filter fights that define the mainstream platforms. It's an 18+ site and behaves like one. This is not where you send a beginner.

• Bring-your-own model- Connect OpenRouter, DeepSeek, OpenAI or other providers with an API key or proxy URL. Chat quality and cost become your call, not the platform's.
• Unfiltered character cards- Community-made characters with detailed personalities, scenarios and example dialogue, tagged and searchable, with no platform filter on outputs.
• JanitorLLM fallback- A free in-house model that works with zero setup. Useful for trying the platform before you configure an external model.
• Creator tools- A full character editor with token counts, permanence settings and public or private publishing for sharing cards with the community.

JanitorLLM on its own earned maybe a 3 from us. Serviceable prose, short memory, and evening rate limits that stall a conversation. Then we connected DeepSeek via OpenRouter, about ten minutes of setup following Janitor's own guide, and the same characters got dramatically sharper. A $5 credit barely moved after a week of heavy use. Memory depends entirely on the model and context size you pay for. The library's depth is real, and so is the jank. No apps, occasional site slowdowns, and quality that swings card to card.

• No content filter on user-created characters, the creative freedom the big platforms won't allow
• Genuinely free. No subscription exists, and the built-in model costs nothing
• Bring-your-own-model design means chat quality can beat any fixed-model rival
• Huge, fast-growing character library with detailed cards and creator tools
• Costs are transparent. You see exactly what the model provider charges, with no markup

• Proxy and API setup is a real wall. Expect to read a help doc before your first good chat
• JanitorLLM, the no-setup default, is mediocre and heavily rate-limited at peak times
• No native mobile apps, and no built-in image generation or voice
• Routing chats through third-party proxies means trusting whoever runs the proxy with your messages
• Character quality swings wildly. The library is community-run with light curation

There are two separate things to trust here. Janitor stores your account and chats on its servers like any platform, so use a throwaway email and no real-name details. The bigger one is the proxy ecosystem. Routing chats through a community reverse proxy hands an unknown third party your messages, and if it's badly configured, your API key too. Use your own API key with an established provider like OpenRouter instead of shared proxies. There's no subscription, so no billing trail from Janitor itself. API charges show up from your model provider.

Best value in unfiltered roleplay for anyone comfortable with a one-time technical setup, held back by proxy plumbing that will lose you if you expected an app-store install. The platform is free, and a few dollars of API credit outperforms most $15/mo subscriptions. Skip it if you want a polished app with images and voice out of the box. That's not what this is. Curious? The site itself costs nothing to try, so use a throwaway email and your own API key rather than a shared proxy.

Is Janitor AI free to use?

Yes. The platform has no subscription at all. The built-in JanitorLLM model is free, though rate-limited at peak hours. Connecting a better external model means paying that provider per use, which on budget models like DeepSeek costs a few dollars for weeks of chat. You can start with zero spend.

How do I set up a proxy on Janitor AI?

In a character chat, open API settings, choose proxy, then paste your provider's endpoint URL, model name and API key. Most people use OpenRouter with a DeepSeek model. Janitor's own help guide walks through it step by step, and it takes about ten minutes the first time.

Does Janitor AI have a content filter?

No. There's no platform-level filter on user-created characters. It's an 18+ site built for unrestricted roleplay, which is the whole reason it exists. One caveat: if you connect an external model, that model provider's own policies still apply to whatever you send through it.

Is Janitor AI safe and private?

The site itself is legitimate, but chats are stored, and community reverse proxies can expose your messages to whoever runs them. Use your own API key with a known provider like OpenRouter, a throwaway email, and no personal identifiers. Done that way, it's reasonably private for an unfiltered platform.

Does Janitor AI have a mobile app?

No, it's web-only. The site works fine in a mobile browser, and you can add it to your home screen, but there's no iOS or Android app and no built-in image generation or voice. If those matter to you, a dedicated companion app fits better.