HomeGuide
Online uncensored LLM: use it via API without installing anything

Updated LLM Sin Censura
To use an online uncensored LLM you don't need a GPU or to download anything: a key, a URL and any OpenAI-compatible client are enough. This guide goes from the key to the first streaming chat and to a desktop client.

Step 1: the key and a 10-second test
An online uncensored LLM works like any cloud API. Sign in with Google or email and password on the API key page, claim the $0.50 trial credit with one click (7 days, no payment method needed, one per person), and copy the key. Then verify that everything responds:
curl https://api.llmsincensura.com/v1/chat/completions \
-H "Authorization: Bearer TU_CLAVE" \
-H "Content-Type: application/json" \
-d '{"model": "uncensored", "messages": [{"role": "user", "content": "Preséntate en una frase."}]}'
If you see a JSON with the response in choices[0].message.content, you have an online model running. It is a single open model hosted by an independent provider: it is not the official API of a large AI company and only handles text.
Step 2: a terminal chat with history
The API does not save the conversation between calls. Each request must carry the previous messages, and for that a growing list is enough. This script keeps the history, shows the response as it arrives and calculates the cost:
import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.llmsincensura.com/v1",
api_key=os.environ["API_KEY"],
)
historial = [{"role": "system", "content": "Eres un narrador de ficción para adultos. Responde en español."}]
gasto = 0.0
while True:
texto = input("tú> ")
if texto in ("salir", "exit"):
break
historial.append({"role": "user", "content": texto})
stream = client.chat.completions.create(
model="uncensored",
messages=historial,
stream=True,
)
respuesta = ""
for chunk in stream:
if chunk.choices and chunk.choices[0].delta.content:
parte = chunk.choices[0].delta.content
respuesta += parte
print(parte, end="", flush=True)
if chunk.usage:
gasto += chunk.usage.prompt_tokens * 0.15 / 1e6
gasto += chunk.usage.completion_tokens * 0.50 / 1e6
print()
historial.append({"role": "assistant", "content": respuesta})
print(f"Gasto de la sesión: ${gasto:.6f}")
Pass the key with export API_KEY=... and run it. Each turn forwards the entire history, so the input grows with each message. A 30-turn session with about 400 tokens per turn ends up sending contexts of about 12,000 tokens; at $0.15 per million, that last input alone costs 12,000 × 0.15 / 1,000,000 = $0.0018.
Step 3: connect it to a chat client
If you prefer a graphical interface, many chat clients accept an OpenAI-compatible endpoint. Applications like Open WebUI or SillyTavern allow you to add a custom connection with a base URL, a key and a model. Menu names vary between versions, so check the documentation for the specific client; the pattern is always the same:
| Client field | Value |
|---|---|
| Connection type | OpenAI compatible / custom |
| Base URL | https://api.llmsincensura.com/v1 |
| API key | Yours |
| Model | uncensored |
If the client allows listing models, use GET /v1/models, which the API does offer. If it doesn't, type the name manually. And if a client asks for the full URL, add /chat/completions at the end.
Some of these clients are installed on your computer or on your own server. This doesn't change the fact that the model runs online: what you install is just the interface.
Step 4: your own small website without exposing the key
Whoever wants a chat on their site usually makes a mistake: calling the API directly from browser JavaScript. That leaves the key visible to anyone. The correct pattern puts a small server between both:
import express from "express";
import OpenAI from "openai";
const app = express();
app.use(express.json());
const client = new OpenAI({
baseURL: "https://api.llmsincensura.com/v1",
apiKey: process.env.API_KEY,
});
app.post("/chat", async (req, res) => {
const r = await client.chat.completions.create({
model: "uncensored",
messages: req.body.messages,
max_tokens: 500,
});
res.json({ texto: r.choices[0].message.content });
});
app.listen(3000);
Add to that server what your case requires: user authentication, a message limit per person, and a check that the history does not grow without limit. With a full key you get 8 concurrent requests and 300 per minute, and that limit applies to all your users at once, not per user.
Best results with a good system prompt
An uncensored model doesn't guess the tone you want. The system prompt is where you set the rules of the game, and it's worth spending five minutes on it:
- Define the role. "You are the narrator of a noir novel set in 1950" works better than "write stories".
- Indicate the format and length. For example: "respond in third person, between 150 and 250 words".
- Set the language. If you want neutral Spanish or voseo, say so explicitly.
- Write what it should not do only if necessary, and positively when you can: "always keep the character's voice" is clearer than a list of prohibitions.
For temperature, start at 0.8 for creative writing and lower to 0.3 if you need more stable responses, for example when extracting data. Change one parameter at a time to know what produced the improvement. And check the usage field after each test: a long system message is paid for in each turn.
Costs, limits, and common errors
Price: $0.15 per million input tokens and $0.50 per million output tokens, with prepaid credit, no subscription. Failed or rejected requests are not charged. With the $0.50 free trial credit, messages of 500 input tokens and 300 output tokens cost 0.000075 + 0.00015 = $0.000225 each, so you get about 2,222 exchanges (0.50 / 0.000225).
Errors you may encounter:
- 429 with
concurrency: too many parallel requests at once. The trial key supports 2; a streaming response occupies its slot until it finishes. - 429 with
rate_limited: you exceeded the requests per minute (60 in the trial). - 429 with
Retry-After: 5: temporary overload, retry after 5 seconds. - The chat forgets what came before: you are not forwarding the history.
- The chat truncates or becomes slow: the history approaches the 100,000-token context window; trim old messages.
When the credit runs out, top ups go from $20 to $500 in crypto (USDT on TRC20 or USDC on Base), with bonuses from $50, according to the pricing page. For advanced options, the documentation.
What type of content is supported
The model is tuned to respond without content rejections for legal adult uses: fiction, roleplay, creative drafts, conversational companionship. The hard limit is that sexual content with minors is not allowed. For more formal integrations, the article on API integration of an uncensored AI expands on the technical side.
And a note on privacy: prompts are not used for training, although they travel to the provider. If you handle information that cannot leave your environment, a local model is the alternative.
More guides
Frequently asked questions
Do I have to install any model?
No. The model is hosted and used via API. You just need a key, the base URL, and an OpenAI-compatible client.
Is there a ready-made web chat?
This service offers the API. To chat with a graphical interface, use a third-party client that accepts an OpenAI-compatible endpoint.
How do I check my remaining balance?
With the X-Balance-USD header of each response, or with GET /v1/balance.
Does it work for generating images?
No, text only.
Your key is one step away
Create an account, copy the key, and change the base URL. That's how easy it is.
Get API key