On Xoul AI the proxy is called a custom engine, and it takes four things: the Generic provider, a base URL, a key and a model name. This page puts an uncensored roleplay build into those fields and keeps the engine within the caps Xoul documents, 875 tokens and 3,500 characters a reply. Every field name here is taken from Xoul's official documentation as read on 24 October 2026.
If you have not met builds with the refusals removed, read AI with no restrictions first and return here to wire one up.
What you need before opening Xoul
- A key from the console. It begins
sk-unr-, appears once, and sits on a prepaid balance you top up from $10 by card, WeChat Pay or crypto. No ID, no KYC. An empty balance still saves, but the first message comes back 402. - A model id, copied, not typed, from the uncensored model list or from the build table further down.
- A Xoul conversation open, since the custom engine form lives under the model selector.
One request confirms the key and shows which ids it reaches. It is free and finishes in a second; put your key in and run it in a terminal.
curl https://api.unrestrictable.ai/v1/models \ -H "Authorization: Bearer sk-unr-live-…"A 200 carrying a list of ids is a working key, and if your chosen id is in the list the Base URL is the only field that can still go wrong. A 401 points at the key, and the body tells you how:
HTTP 401{"error":{"message":"No API key was sent.","type":"authentication_error","param":null,"code":"invalid_api_key"}}HTTP 401{"error":{"message":"The API key is not valid.","type":"authentication_error","param":null,"code":"invalid_api_key"}}To try a model with no setup at all, the same ids answer in the browser chat, three free turns a day and no account.
Set up the Xoul custom engine, step by step
- In the conversation, find the model selector and add a custom engine profile beneath it. Xoul's announcement gives this path; its Custom Engine (Proxy) page shows the fields without the clicks.
- Give it an Engine Name, say Unrestrictable. Xoul stores the engine under that name, so next time it is one click.
- Choose Generic as the Provider. A Base URL field opens; put
https://api.unrestrictable.ai/v1in it. - Drop the key into API Key. Xoul's doc says keep it private, and it is right: whoever has it can run down your balance.
- Under Model Name, pick the id from the list, or type and then select it,
cydonia-24b-v4.3for instance. - If you like, add a System Prompt and set Max Context Length to the number in the build table below.
- Put Max Tokens at 875 or under. For the first message leave Temperature, Top P, Seed, Top K, Min P and Repetition Penalty where they are.
- Save, then send one line. If a reply comes back, the URL, the key and the id all check out.
The six fields, with their values
Here is the whole form for this API in one place, using the labels from Xoul's Custom Engine (Proxy) doc.
- Engine Name
- UnrestrictableYour choice; Xoul keeps it for next time
- Provider
- GenericReveals the Base URL field
- Base URL
- https://api.unrestrictable.ai/v1Ends in /v1
- API Key
- sk-unr-live-…From the console; displayed one time
- Model Name
- cydonia-24b-v4.3Any id on the Models page
- Max Tokens
- 875The ceiling in Xoul's doc
Xoul's doc is silent on the shape of a Generic Base URL. A URL that ends in /v1 is what every OpenAI-compatible client assumes, and it is the form the docs page gives for each client of this API.
Three limits Xoul puts on a proxy
These come from Xoul's documentation, not from the API, and the engine has to fit inside them.
- Max Tokens stays at or below 875. That is the reply length Xoul recommends. Give a model 875 tokens and it stops there, even in the middle of a sentence.
- A reply holds at most 3,500 characters. The doc allows that a proxy may send more, but once you edit a reply over 3,500 characters it will not save. At roughly four characters to a token in English, the two caps describe the same length.
- No Response Styles, no creator system prompts. Xoul withholds them from proxies because, as the doc puts it, a proxy could leak them. Whatever instructions you need go in the engine's own System Prompt field instead.
Picking an uncensored build for Xoul
The catalog tags each of these three as roleplay, and each speaks the OpenAI wire format. Price and window are pulled from the live catalog as the page renders.
| Wire formats | |||||
|---|---|---|---|---|---|
| Cydonia 24B v4.3 | Roleplay | $0.484 | $0.60 | 131K | |
| Qwen3.8 27B Antislop | Roleplay | $0.35 | $2.10 | 262K |
| Model id | Catalog window | Max Context Length to type |
|---|---|---|
cydonia-24b-v4.3 | 131,072 tokens | 32768 |
qwen3.8-27b-antislop | 262,144 tokens | 32768 |
Do not go above the window. Xoul's doc adds that a lower context limit makes each call cheaper, because the full context travels as input with every message. qwen3.8-27b-antislop belongs to the Qwen 3.8 tunes; its siblings are on the uncensored Qwen page.
The bill for one evening of Xoul chat
Take 200 messages in a single session. Each reply is charged as if it hit the 875-token ceiling, so these are ceilings for each context size; actual replies tend to run shorter. In the short column every message carries about 4,000 tokens of character card and history, in the long column 16,000.
| Model | 200 messages at 4,000 context | 200 messages at 16,000 context |
|---|---|---|
| Cydonia 24B v4.3 | $0.49 | $1.65 |
| Qwen3.8 27B Antislop | $0.65 | $1.49 |
The input side is the bulk of it: 200 × 4,000 comes to 800,000 input tokens, against no more than 175,000 tokens out. Trim Max Context Length and the bill drops further than it would from switching models. All per-token rates are on the pricing page.
How a Xoul custom engine works
Xoul AI lets you build AI characters, scenarios and whole worlds, then chat with them. By default those chats run on models Xoul supplies.
Xoul's word for a proxy is custom engine: the chat goes to a model on an API of your choosing, paid for with your own key. The feature was announced in Xoul's official documentation, where the custom engine profile is added from a conversation, below the model selector. Its fields are set out on the Custom Engine (Proxy) page. Both pages were read on 24 October 2026.
When the Xoul engine returns an error
401 is the key
"No API key was sent." means nothing is in the API Key field. "The API key is not valid." means the key is wrong, truncated or revoked. Make a new one in the console and paste the whole string.
404 model_not_found is the model id
The catalog has no such Model Name: a misspelling, a display name where an id belongs, or a build that has been retired. Rerun the request above and copy the id straight from the output.
402 is the balance
Nothing is left on the prepaid balance. Add funds in the console. A 402 request costs nothing.
CORS in the browser
Every origin may call the API, Authorization header included (confirmed 24 October 2026). From Xoul or from your own browser, there is no setting to flip.
Replies that stop mid-sentence
Not a fault. The reply hit Max Tokens. Raise it as far as 875, or tell the model in the System Prompt to keep replies short.
Once a day the real-model check sends a fixed prompt to every id, so the model answering under a name is the model that name promises.
Questions
Xoul publishes its own plans at xoul.ai. With a custom engine the model use is billed by the API you plug in: here that is per token, drawn from a prepaid balance that opens at $10. See per-token prices.
A proxy routes your chat to a model on a different API rather than the app’s own. In Xoul it is the custom engine: base URL, key, model name. Xoul still holds your characters and history; the API does the generating.
Turn it on when you want a model Xoul does not carry, an uncensored roleplay build for example, charged to your own key. Turn it off to get Xoul’s built-in models back along with Response Styles and creator system prompts, which Xoul disables whenever a proxy is active.
875 at most, which is the figure in Xoul’s doc. Replies are capped at 3,500 characters, and anything longer cannot be saved once edited.
The Generic provider accepts any id that speaks the OpenAI wire format. For characters, begin with one the catalog tags roleplay; three are in the table above with windows and prices. See the uncensored model list.
No. These builds are abliterated or uncensored fine-tunes: the refusals are gone from the weights themselves, so nothing has to be pasted into the System Prompt to get there. See AI with no restrictions, explained.
Every model above is on the uncensored model list with its live price per million tokens. Per-token pricing sets the whole catalog side by side, and the family directory groups the builds by base model.