Uncensored API FAQ

The basics

What is Uncensored API in one sentence?

It is a paid, text-only chat completions endpoint that serves a single model, id uncensored, and does not refuse lawful adult content, fiction or controversial subjects.

Which URL do I call?

Use https://api.uncensoredapis.com/v1 as the base. Chat requests go to POST /v1/chat/completions and the model list lives at GET /v1/models. Any other path answers 404.

Do I have to learn a new SDK?

No. The request and response shapes follow the common chat completions convention, so the official Python and Node packages work once you change the base URL and key. Plain curl works too, as the docs show.

Is there a way to try it before paying?

Yes. A new account receives $0.50 of trial credit, valid for 7 days, and no payment details are requested at sign-up. Register with an email and password on the key page and the key appears immediately.

Can I choose between several models?

No. There is exactly one model, and its id is uncensored. A call to the models route returns that single entry, so there is nothing to select or switch between.

Who is the service meant for?

Developers building products for adults, including fiction tools, roleplay apps and research assistants. You are responsible for your own age checks and content rules in whatever you ship on top.

Keys and access

How many keys does an account get?

One. If you need a new value, regenerate it from the account; the previous key stops working at the same moment, so deploy the replacement right away.

Where should the key live?

Keep it in an environment variable or a secret manager on a server you control. Browser JavaScript and mobile binaries can be inspected, so route client traffic through your own backend instead of embedding the key.

How is the key sent?

In an Authorization: Bearer header on every request. A missing or wrong value returns 401 with a JSON error object.

I want an uncensored API key for a team. Can we share one?

You can, but every process shares the same 300 requests per minute allowance and the same balance. Put a small gateway in front of the key and give each teammate access to the gateway rather than the key itself.

Limits and pricing

How big can a prompt be?

The context window is 100,000 tokens, shared by prompt and completion. The body of a single request may not pass 8 MB. A request whose prompt plus max_tokens would exceed the window is rejected with a 400.

How long can a reply be?

The default is 2048 tokens. You can raise max_tokens to at most 16,000 per request. When an answer stops because of the cap, finish_reason reads length.

What does it cost?

Input tokens are $0.25 per million and output tokens are $1.00 per million. Balance is prepaid, there is no subscription, and the balance never expires. See the pricing page for the table.

Can you give a quick estimate?

Assume a chat turn with 1,500 prompt tokens and 300 completion tokens, purely as an illustration. That is $0.000375 for input plus $0.0003 for output, or about $0.000675 per turn, so $1 buys roughly 1,480 such turns.

What happens when I send too many requests?

Each key may make 300 requests per minute. Beyond that you get a 429, and the right response is to slow down and retry with a delay.

Integration questions

Which sampling parameters can I set?

Standard ones such as temperature, top_p and stop are passed through to generation. Set them in the request body exactly as you would elsewhere.

Does the API remember earlier turns for me?

No. Each request is independent, so you resend the conversation in messages every time. Trim the oldest turns when the total nears the 100,000-token window.

Can I stream tokens to a browser?

Yes, send stream: true and read the server-sent events. Do it from your backend and relay the text to the page, so the key never reaches the client.

Is a system message supported?

Yes. Put a message with the system role first in the array to set tone, format and boundaries for the conversation.

What should I check first when output looks cut off?

Read finish_reason. A value of length means the cap was reached, so raise max_tokens or shorten the prompt. The reference lists all values.

Balance and billing

How do I add funds?

Top up your prepaid balance from your account. There is nothing to subscribe to, and unused balance stays available because it never expires.

What happens when the trial credit runs out?

The trial credit is valid for 7 days. After it is spent or expired, requests return 402 with the code no_credit until you top up the balance.

Can I see what a request cost?

Each response includes a usage block with prompt, completion and total tokens. Multiply by the per-million rates to get the charge, and keep your own tally if you want a ledger.

Are failed requests free to retry?

Retrying is allowed, but keep it bounded. Use a short backoff for 429 and 503 and never loop on 400, 401, 402 or 403, since they will not fix themselves. The errors playbook has a ready-made policy.

Content and behavior

What will the model not do?

Lawful adult content is not refused, and neither are fiction or contentious topics. The firm exception is any sexual content involving minors, which is blocked with a 403 every time, including in stories and roleplay. The service is for adults aged 18 and over.

Are my prompts used to train a model?

No. Prompts are not used for training.

How does this differ from routing through an aggregator?

An aggregator forwards you to many models, each with its own rules. This service is one model behind one policy. The uncensored OpenRouter guide walks through that comparison in depth.

Can it see images or make embeddings?

No. It handles text only, with no embeddings, images, audio, video or fine-tuning.

Troubleshooting

I get 402 with no_credit. What now?

Your balance is used up or the trial lapsed. Top up the prepaid balance and send the request again.

A request returns 503 upstream_busy.

It is temporary. Wait a few seconds and retry, using a growing delay and some randomness if you run many workers.

My stream ends and I cannot find the token count.

With streaming on, a final chunk containing usage is added automatically, right before data: [DONE]. Its choices list is empty, so check for that before indexing.

Do tools and function calling work?

Yes, in the standard format: pass a tools array and look for tool_calls in the reply.