What "uncensored" means here
This refers to a single open-weight model operated by an independent provider and tuned to not refuse by principle. It answers questions about fiction, roleplay, dark themes in stories, or sensitive factual topics that a carefully configured assistant would refuse. It is intended for lawful use by adults.
It is neither the official API of a major vendor nor a router over multiple providers. And it is text only: no images, no audio, no video. Anyone who wants to use such a model therefore only needs two things, a key and a client.
Step 1: Set up access
- Sign in with Google or with email and password.
- Generate your key on the Get API key page.
- Claim your $0.50 free trial credit with one click there. It is valid for 7 days, limited to one per person, and requires no payment details.
- Note the base URL
https://api.kiohnezensurapi.com/v1and the model nameuncensored.
The test key allows 2 parallel requests and 60 per minute. That is enough for personal use. A full key after a top up allows 8 parallel requests and 300 per minute.
Way A: A chat client with a custom endpoint
Many chat interfaces and roleplay programs can use a custom OpenAI-compatible endpoint. Enter three things there:
- API type or source: OpenAI-compatible (the label varies by program)
- Address or base URL:
https://api.kiohnezensurapi.com/v1 - Key and model: Your key, model name
uncensored
Labels and menus change between versions. Therefore, check your client's help for where to enter a custom endpoint. Two settings are important: Do not set the context window higher than you really need (the API context window covers 100,000 tokens for prompt and response), and enable streaming so responses appear piece by piece.
After entering, verify with your client's connection function or with GET /v1/models that everything arrives.
Way B: Your own code
With Python, a short script is enough. The system prompt determines how the model presents itself.
import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.kiohnezensurapi.com/v1",
api_key=os.environ["KI_KEY"],
)
system = (
"Du bist ein erfahrener Autor für Kriminalromane. "
"Du schreibst düster und konkret, ohne zu moralisieren. "
"Antworte auf Deutsch."
)
r = client.chat.completions.create(
model="uncensored",
temperature=0.9,
messages=[
{"role": "system", "content": system},
{"role": "user", "content": "Beschreibe den Tatort in einer verregneten Hafenstadt."},
],
)
print(r.choices[0].message.content)The same with curl, if you just want to test:
curl https://api.kiohnezensurapi.com/v1/chat/completions \
-H "Authorization: Bearer $KI_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "uncensored", "messages": [{"role": "user", "content": "Hallo."}]}'Further parameters, streaming, and tool calls are explained in the documentation.
The system prompt makes the difference
A model without refusal reflex is not automatically a good model for your task. Quality depends heavily on what you tell it. Proven building blocks:
- Role and goal: Who is speaking, with what intent?
- Tone: matter-of-fact, ironic, literary.
- Format: length, paragraphs, dialogue form.
- Taboos: what you do not want, stated calmly and concretely.
- Example: two sentences in the desired style often work better than ten instructions.
Always change only one thing and compare. With seed and low temperature, comparisons become reproducible. For creative tasks, you may raise the temperature.
What people use the model for
The most common applications are unremarkable and easy to describe:
- Fiction and screenplay drafts: antagonists, violence in crime or war novels, morally grey characters, without the model launching into a lecture mid-sentence.
- Role-playing and companion apps for adults: characters with a fixed personality that remain consistent across many messages.
- Research on sensitive topics: security research, history, medicine, and law as a starting point, never as a replacement for experts.
- Text processing without sugarcoating: summarizing or rewriting material that a cautious model would refuse.
For each of these cases: describe in the system prompt what you expect, and critically evaluate the first results. If the output seems too tame or too verbose, change the role and format first before adjusting the temperature.
If no response arrives
If the client does not respond, check in sequence: Does the base URL end with /v1? Is the key inserted without spaces? Is the model set to uncensored? Is there still credit? The header X-Balance-USD in every response shows this.
History, costs, and limits
The API remembers nothing. Your client or code sends the entire history with every message. This matters for costs. An example with $0.15 per 1M input and $0.50 per 1M output: A message with 6,000 tokens of history and 500 tokens of response costs 6,000 / 1,000,000 × $0.15 = $0.0009 plus 500 / 1,000,000 × $0.50 = $0.00025, totaling $0.00115. One hundred such messages cost $0.115.
To top up: any whole amount from $20 to $500, payment with USDT on TRC20 or USDC on Base, with a +5% bonus from $50, +10% from $100, and +15% from $200. The credit never expires, and rejected or failed requests are not charged. Details on the pricing page.
If you receive HTTP 429, read the error code: rate_limited means too many requests per minute, concurrency means too many concurrent requests. A streaming request occupies a slot until the stream ends.
Limits that remain
A model without refusals does not mean everything is allowed. Three points are binding:
- Minors: Sexual content involving minors is excluded without exception.
- Law: The use is intended for lawful purposes. The model does not replace a legal review of your project.
- Responsibility for results: A model can state falsehoods. Check facts before passing them on, especially regarding health, law, and finance.
Your prompts are not used for training. An overview of the topic is provided in the article AI without restrictions.

