EN ▾
Get API key
OpenAI-compatible uncensored API for developers.Updated

Uncensored LLM API for Developers: How It Works

An uncensored llm api for developers is a hosted model endpoint that answers lawful adult requests instead of refusing them by default. This page explains what that means in practice, where one fits in an application, and what to check before you build on it.

Abstract terminal streams passing through an open gateway, representing an unrestricted text API for developers

What the phrase actually means

Most hosted language models ship with a layer of refusal behaviour. That layer is a sensible default for a general assistant, but it gets in the way of many legitimate products: fiction tools that need villains with menace, roleplay and companion apps for adults, moderation research that has to look at the content it classifies, and red-team test harnesses. An uncensored model is one that has been tuned not to refuse these requests, and an uncensored API is simply that model behind an HTTP endpoint.

The word "uncensored" does not mean "no rules". Our service has one hard limit: no sexual content involving minors, ever. Beyond that, the model is tuned to answer without content refusals for lawful adult use. The responsibility for what your product does with the output, and for following the law where you operate, stays with you.

It also helps to be clear about what the model is not. It is one open-weight model run by an independent provider. It is not the official API of Meta, Mistral, Qwen, DeepSeek, OpenAI, xAI, Google or Anthropic, and it is not a router that picks among vendors.

Where it fits in an application

Developers usually reach for this kind of endpoint in one of four situations.

  • Creative writing and game dialogue. Story engines, interactive fiction and NPC conversation break when the model lectures the player. A model that stays in character keeps the experience intact.
  • Companion and roleplay apps for adults. Persona chat depends on long sessions and consistent tone. The 100,000-token window helps here, since a long conversation plus a character sheet fits without aggressive truncation.
  • Data and content pipelines. Classifying, summarising or rewriting user-generated text that contains rough language, without the model refusing half the rows.
  • Testing and research. Comparing how a filter behaves against raw model output, or generating adversarial examples for a safety classifier you are building.

In all four, the model is one component. You will still write your own system prompts, your own age gating and your own moderation for what you publish. The API gives you generation, not product policy.

What you get technically

The service exposes two methods: POST /v1/chat/completions and GET /v1/models. Auth is a bearer token in the Authorization header. The model ID is uncensored. Because the protocol matches OpenAI's, any OpenAI SDK or compatible client works once you change the base URL and the key.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.uncensoredllmapi.com/v1",
    api_key="YOUR_API_KEY",
)

resp = client.chat.completions.create(
    model="uncensored",
    messages=[
        {"role": "system", "content": "You write gritty fantasy dialogue."},
        {"role": "user", "content": "A mercenary refuses a noble's offer. Three lines."},
    ],
    temperature=0.9,
)
print(resp.choices[0].message.content)

Supported features include streaming over SSE with usage reported at the end, tool calling (tools, tool_choice, streamed tool_calls and results as role: "tool"), and JSON mode through response_format. Sampling settings such as temperature, top_p, stop, seed and the penalties are passed through. The full request reference is in the API documentation.

Limits are 8 concurrent requests and 300 requests per minute on a full key, and 2 concurrent and 60 per minute on a trial key. The maximum body is 8 MB. When you exceed a limit you get HTTP 429, with the error code rate_limited or concurrency.

What it does not do

Being straight about the gaps saves you an integration cycle. The service is text only. There is no image, audio or video generation, no embeddings endpoint and no fine-tuning. There is one model, so you cannot choose between sizes or switch to a different vendor's model by changing a string. If your pipeline needs retrieval, you will need an embeddings source elsewhere and then pass the retrieved text into this API as ordinary context.

If you already have an application that picks from a catalogue of many models, a single-model endpoint is a narrower tool. For the operational side, read the production checklist for NSFW integration.

The model is also not guaranteed to be right about facts. Removing refusals does not add knowledge. Treat output as draft text, and verify anything that matters.

Sizing the cost

Billing is per token: $0.15 per 1M input tokens and $0.50 per 1M output tokens. You prepay credit, nothing renews, and credit never expires. Failed or refused requests are not charged. Prompts are not used for training.

Here are two worked examples.

A chat turn in a companion app. Suppose each turn sends 3,000 tokens of history and persona, and the model replies with 250 tokens. Input costs 3,000 x $0.15 / 1,000,000 = $0.00045. Output costs 250 x $0.50 / 1,000,000 = $0.000125. One turn is $0.000575, so 1,000 turns cost $0.575.

A batch rewrite job. Take 20,000 text snippets of 400 input tokens each, with 300 tokens returned for each. Input is 8,000,000 tokens, which is 8 x $0.15 = $1.20. Output is 6,000,000 tokens, which is 6 x $0.50 = $3.00. The job costs $4.20 in total.

Output tokens cost more than three times as much as input tokens, so limiting reply length is the biggest saving on generation-heavy work. Long conversations also grow the input side every turn, which is a reason to summarise old history rather than resend it. The pricing page lists the current rates.

Paying and getting started

New users can claim $0.50 of trial credit with one click on the API key page after signing up with Google or with email and password. It is valid for 7 days, limited to one per person, and needs no payment. Run a few real prompts from your own use case and check that the tone and format fit.

When you move to production, top up any whole amount from $20 to $500. A bonus applies at larger sizes: +5% from $50, +10% from $100 and +15% from $200. This site accepts crypto only, using USDT on TRC20 or USDC on Base. If you have never paid in crypto, plan a little time for your first transfer and top up before you need the credit.

Create your key on the API key page, then point your existing client at the base URL.

A short evaluation plan

  1. Write ten prompts that your product will really send, including a few that other models refuse.
  2. Run them with the trial credit at your intended temperature and note refusals, tone and length.
  3. Test streaming and, if you rely on them, tool calls and JSON mode.
  4. Estimate monthly tokens from the examples above and multiply by the two rates.
  5. Decide on your own moderation and age gating before launch.

If those five steps go well, you have what you need to decide whether this endpoint belongs in your stack.

Questions and answers

Is an uncensored LLM API legal to use?

The service is built for lawful adult use, and you are responsible for your product's content and for following the laws where you and your users are. The service forbids any sexual content involving minors.

Can I use the OpenAI SDK?

Yes. Set the base URL to https://api.uncensoredllmapi.com/v1, use your key as the bearer token and set the model to uncensored.

Does it support function calling?

Yes. Tool calling with tools, tool_choice and streamed tool_calls is supported, and JSON mode is available through response_format.

Is there a free tier?

No. New users can claim $0.50 of trial credit on the API key page, valid for 7 days, one per person, with no payment needed.

Your key is one form away

Create an account, copy the key, change the base URL. That is the whole setup.