An uncensored AI API is a model endpoint you call from your own code or front end, serving a model that writes adult and dark content without refusing it. Most of them use OpenAI's Chat Completions format, so code that already talks to a model needs a new base URL and a new key, and little else. Hemmingway Unlocked has one at https://wise.hemmingway.io/v1.
Two things to know before you build. It is for adults, and that includes the people who use what you build. And uncensored does not mean the model will write anything at all: every service has a list of things it refuses, and your app inherits that list. Read it first.
Who needs one
- People with their own front end. You play or write in SillyTavern or a similar tool and want a model behind it that does not stop the scene.
- Developers of adult fiction tools. Interactive stories, a drafting tool for authors of horror or dark romance, a companion app for adults.
- Writers with a script. You draft or rewrite in bulk, and the model you use now keeps refusing the dark chapters.
If your product is for a general audience, you do not want an uncensored model. Use an ordinary one.
Three ways to get one
| Route | What it is | Good for | Watch out for |
|---|---|---|---|
| Host an open model yourself | An uncensored open model on your own hardware, behind a local server | Privacy and full control | You need the hardware, and the model's licence sets what you may do with it |
| A hosted uncensored API | A company runs the model and gives you a key | No hardware, a stable address, rules in writing | It is paid, and your requests pass through its servers |
| A general-purpose API | The large assistants' own APIs | Everything that is not adult content | They are built for a general audience |
Hosting it yourself. vLLM, Ollama, LM Studio and KoboldCpp can each serve a model on your own machine through an OpenAI-compatible endpoint, according to their own docs, so your code does not care whether the model is local or hosted. The work is in choosing the model and owning enough memory to run it. The guide to uncensored LLMs covers how those models are made and how to size one.
A hosted one. This is the quick route: a key, an address, and no machine to look after. The cost is trust. You are relying on the company's rules and on what it says it keeps.
A general-purpose API. On 27 September 2026 we sent the same adult roleplay request to ChatGPT, Claude and Gemini in their chat apps, and each refused. We did not test their APIs, so we say nothing about those here. The test is written up here.
What to check before you build on one
- The rules, in writing. Find the list of what the model will not write. Your users will meet those limits halfway through a scene, so you should know them before they do.
- Who it is for. An adult API is for adults. Your own product needs its own age check and its own terms.
- What is kept. Ask whether requests and replies are stored, for how long, and whether they are used for training.
- The format. If it takes OpenAI's Chat Completions body, most tools will work. Still test streaming and anything else you rely on.
- What a refusal looks like. Find out whether a refused request comes back as an error or as ordinary text, so your code can show the user something sensible.
- How a key is paid for. A plan, prepaid credit or pay as you go. Check what happens when it runs out.
- Whether you can leave. If the weights are open, you can move to your own hardware later. If they are not, you cannot.
Hemmingway Unlocked's API
Hemmingway Unlocked is the adults-only side of Hemmingway: the same app, with an uncensored model. Its API is OpenAI-compatible.
- Base URL:
https://wise.hemmingway.io/v1 - Key: you create it inside Unlocked, in the account menu under API keys. A key needs a membership. The free trial is for the app, not the API.
- Request: the same format as Hemmingway's regular API, which the API docs set out: OpenAI's Chat Completions body, sent with your key as a Bearer token in the
Authorizationheader.
So if your code already calls Hemmingway or OpenAI through OpenAI's SDK, the change is the base URL and the key. In Python that is OpenAI(base_url="https://wise.hemmingway.io/v1", api_key=...), and the rest of the call stays as it was. Send one plain request first, then test the extras you depend on before you build on them.
What the model writes: sex between adults, explicit and in full, and dark, violent stories, with no lecture and no refusal on grounds of taste. What it never writes, however it is asked: anything sexual involving anyone under 18, a real person, animals, family or anyone who does not consent; and, in any kind of text, torture, help hurting or harassing a real person, weapons, making drugs, harmful software, or help with self-harm. The list is in our terms.
Each message passes through our servers to be answered and is not kept, and chats are never trained on. The model writes text only. It does not make images. Its weights are not out yet. We plan to open-source it soon.
Open Hemmingway UnlockedTry Hemmingway
Using it from a front end
Any tool that accepts a custom OpenAI-style address can use the API, SillyTavern included. In SillyTavern you choose the Chat Completion API type, pick the custom OpenAI-compatible source, and enter the address and your key. The SillyTavern API guide goes through it step by step. For what the phrase "OpenAI-compatible" does and does not promise, see OpenAI-compatible APIs.
Your side of it
An uncensored model does what it is asked more often, so more of the care is yours.
- Check ages. Confirm your users are adults before they reach the model.
- Keep the key on your server. A key placed in a web page or an app you hand out can be copied by anyone who looks.
- Write your own rules. Say what your product is for and what it is not for, in plain words your users can find.
- Keep real people out of it. Sexual or violent scenes about a real person are where fiction stops being fiction.
- Give people a way to report. If your users see the model's output, they need somewhere to send a bad answer.
Where it does not fit
- Nothing may leave your machines. Then host an open model yourself. Unlocked's model cannot be downloaded yet.
- You want pictures. It writes text.
- Your product is for everyone. Use Hemmingway's regular API at
https://hemmingway.io/v1. Keys and prices for that one are on the API platform.
Common questions
Is there an uncensored AI API?
Yes. Hemmingway Unlocked has one at https://wise.hemmingway.io/v1, in OpenAI's format, for adults. You can also serve an uncensored open model on your own hardware and call it the same way.
Is the uncensored AI API free?
Hemmingway Unlocked is free to try in the app, with no card, but an API key needs a membership. Hosting an open model yourself needs no subscription, only the hardware.
Is Hemmingway Unlocked's API OpenAI-compatible?
Yes. It takes OpenAI's Chat Completions format, so OpenAI's SDKs and tools that accept a custom OpenAI-style address can use it. You change the base URL and the key.
Does an uncensored API have any rules?
Yes, every one does. Hemmingway Unlocked never writes anything sexual involving minors, real people, animals, family or anyone who does not consent, and never helps with hurting a real person, weapons, making drugs or self-harm. Read a service's list before you build on it, because your app inherits it.
Can I use an uncensored API with SillyTavern?
Yes, if the API is OpenAI-compatible. In SillyTavern, choose Chat Completion, pick the custom OpenAI-compatible source, and enter the address and key. For Hemmingway Unlocked the address is https://wise.hemmingway.io/v1.
Can I self-host an uncensored model instead of using an API?
Yes, with an open model and enough memory to run it. It is the most private route and the most work. Hemmingway Unlocked's own model is not published yet, so that one is only available hosted for now.