Unfiltered AI API: frequently asked questions
This page collects the practical questions developers raise before and after integrating Unfiltered AI. It is grouped by layer rather than alphabet: the product, access and billing, request behaviour, compatibility, content policy and errors. Each answer sticks to what the service does today, so you can plan around it with confidence.

UpdatedUnfiltered AI
Key points
- Dedicated inference providers like Unfiltered AI offer consistent uncensored behavior on a single model, avoiding the randomness of aggregator routing.
- Aggregators like OpenRouter provide access to multiple models but may vary in how strictly they apply content filters depending on the specific model version.
- Unfiltered AI uses an open-weight model on dedicated hardware with a 100k context window, priced at $0.15/$0.50 per million tokens.
- For strict NSFW or roleplay consistency, a dedicated uncensored LLM API eliminates the need for complex prompt engineering to bypass generic filters.
What the service is, and what it is not
Questions tend to cluster, so this page is organised by the layer of the stack they concern: first the product itself, then access and billing, then request behaviour, then content and privacy, then fixes. If your question sits between two layers, read both.
What exactly is Unfiltered AI?
A hosted text-generation API that follows the OpenAI chat-completions format. You send a list of messages and receive generated text, either all at once or streamed. It has a single model with the id uncensored.
What can it not do?
It handles text only. There are no embeddings, image, audio or video endpoints, and no fine-tuning. Two endpoints exist: POST /v1/chat/completions and GET /v1/models.
How is this different from the comparison article?
The comparison helps you pick among options. This page assumes you are already considering this service and answers practical questions about it.
Access and billing
Access is deliberately simple: one account, one key, one prepaid balance. Nothing renews on a schedule.
How do I start?
Create an account on the key page with an email and password. A key is displayed immediately, and new accounts include $0.50 of credit valid for 7 days, with no payment details needed.
How is usage charged?
Per token: $0.15 for every million input tokens and $0.50 for every million output tokens. You top up a prepaid balance and it is drawn down as you use it.
Is there a monthly plan?
No subscription exists, and a prepaid balance does not expire. You pay only for tokens you actually send and receive.
Can I rotate my key?
Yes. Regenerating issues a new key and the previous one is invalid straight away, so deploy the new value promptly. Each account has a single key at a time.
For the rate card itself, open pricing.
How requests behave
These are the numbers and mechanisms that shape your integration. Think of them as the contract between your code and the endpoint.
| Property | Value |
|---|---|
| Context window | 100,000 tokens, prompt plus completion |
Default max_tokens | Optional. If not set, the response can run to the end of the window |
Maximum max_tokens | Up to the end of the 100,000-token window |
| Request body | Up to 8 MB |
| Rate limit | 300 requests per minute, per key |
| Streaming | Server-sent events with stream: true |
Does streaming report token usage?
Yes. A closing usage chunk is added automatically after the last content chunk, so you can total tokens without a second call.
Which sampling parameters work?
temperature, top_p and stop are passed through in the standard way.
Are tools supported?
Function calling uses the OpenAI tools format. Declare functions, handle returned tool calls and reply with tool messages; the game dialogue article shows a full loop.
Using existing OpenAI client code
Because the interface matches, most projects need two changes: the base URL and the key. Everything else, including message arrays and streaming iterators, stays as it was.
import os
from openai import OpenAI
client = OpenAI(base_url="https://api.unfilteredaiapi.com/v1", api_key=os.environ["API_KEY"])
reply = client.chat.completions.create(
model="uncensored",
messages=[{"role": "user", "content": "Describe a rainy railway platform in one paragraph."}],
)
print(reply.choices[0].message.content)Libraries that accept a custom base URL, such as LangChain or LlamaIndex, can be pointed at the same address. Remember that the only model id is uncensored; asking for any other name will not select a different engine. To confirm the connection and see the model list, run this one-liner.
curl https://api.unfilteredaiapi.com/v1/models \
-H "Authorization: Bearer $API_KEY"If you are building a chat product around the API, the FastAPI backend is a reasonable starting shape, and the memory article covers keeping long sessions inside the window.
Content policy and privacy
The service is built for adults, and the rules are short enough to state completely.
What is allowed?
Lawful adult content, fiction and controversial topics are not refused. The model does not decline a request merely because the subject is dark or explicit between adults.
What is always blocked?
Sexual content involving minors, in any framing, including fiction and roleplay. Such requests return a 403 with the code content_blocked.
Who may use it?
Only people aged 18 and over. If you build a product on top, you are responsible for gating your own audience.
Are prompts used for training?
No. Prompts are not used for training.
Integration questions that come up second
After the first successful call, a different set of questions appears. These are the ones that recur.
Where should the API key live?
On a server you control, in an environment variable. Browser code and mobile apps can be inspected, so route their requests through your own backend.
Do I need to send the whole conversation each time?
Yes. The endpoint is stateless, so each request must carry the messages you want the model to see. Trim or summarise once history approaches the window.
Can several services share one key?
They can, but they share the 300-requests-a-minute allowance, so coordinate their throttles. Regenerating the key invalidates it for all of them at once.
What happens to a request that is cut off by max_tokens?
The reply simply ends where the limit falls. Raise the value while keeping prompt plus completion under 100,000 tokens.
Is there anything to configure for streaming?
Set stream: true and read server-sent events. Disable response buffering in any proxy between the API and your users, or the text will arrive in bursts.
Errors and what to do about them
Every failure returns JSON of the form {"error":{"code":...,"message":...}}. Match the status first, then read the message.
- 400: the request is malformed or too large, for example a prompt beyond 100,000 tokens or an invalid
max_tokensvalue. - 401: the key is missing or invalid, often because it was regenerated.
- 402 (
no_credit): the balance is used up or the trial has expired. - 403 (
content_blocked): the request fell into the always-blocked category. - 404: the path does not exist; only the two endpoints above are served.
- 429: you passed 300 requests in a minute on that key.
- 503 (
upstream_busy): a short-lived overload.
Which errors are worth retrying?
Only 429 and 503. Wait a few seconds, ideally with exponential backoff, then resend. The rest require a change in your request, key or balance.
My replies end abruptly. Is that an error?
Usually not. If max_tokens is not set, the response can run to the end of the window.
Planning a deployment
A last short checklist for people moving from experiment to product. Keep the key on a server and never in browser code. Add a per-user cooldown, since all of your users share one rate limit. Log token usage from the response so costs are visible on your side. Decide how your app reacts to a 402 before the day it happens. State clearly in your interface that the experience is for adults.
Cost is easiest to reason about with a worked case. Assuming a chat turn sends 2,000 input tokens and receives 300 back, the price is 2,000 x $0.15 / 1,000,000 plus 300 x $0.50 / 1,000,000, which is $0.0005 plus $0.0003, or $0.0008. A thousand such turns cost about $0.80. Real prompts differ, so measure with the usage figures your own traffic returns, and revisit the estimate once a week at first.
Questions and answers
How much does Unfiltered AI cost?
Input tokens cost $0.15 per million and output tokens $0.50 per million, drawn from a prepaid balance that never expires and has no subscription.
What are the main request limits?
The context window is 100,000 tokens including the reply, max_tokens is optional and has no separate cap, bodies may be 8 MB, and each key may make 300 requests a minute.
Can I reuse my existing OpenAI client code?
Yes for chat completions. Change the base URL to https://api.unfilteredaiapi.com/v1, supply your key and use the model id uncensored.
What content is blocked?
Sexual content involving minors, including fiction and roleplay, returns 403 content_blocked. Lawful adult content and fiction are not refused, and the service is for adults only.
Is there a free trial?
New accounts receive $0.50 of credit for 7 days with no payment details, and the key is shown as soon as you register.
Your key is one form away
Create an account, copy the key, change the base URL. That is the whole setup.