RisuAI custom API: connect an uncensored roleplay model

Fill the four RisuAI custom API fields (URL, Key/Password, Request Model, Format) with an uncensored roleplay build, then see what a session costs.

Steps checked October 24, 2026 · prices as of November 16, 2026

Connecting RisuAI to this API means choosing one model entry and filling four boxes: URL, Key/Password, Request Model and Format. Below you get the values for an uncensored roleplay build, a sampler preset, and the price of a session on four builds, computed from the live catalog. Field names were taken from RisuAI's own source on 24 October 2026, so they match today's app and not the "Reverse Proxy" wording of older posts.

New to builds that had their refusals removed? Read AI with no restrictions first, then return here to wire one up.

What you need before you start

  • An API key from the console. Keys begin with sk-unr- and appear once, so copy yours into a safe place right away. Credit is prepaid, from $10 by card, WeChat Pay or crypto. An empty key still passes the setup; its first chat message comes back as a 402.
  • A model id, copied (not typed) from the uncensored model list or from the table further down.
  • RisuAI, either the web build at risuai.net or a desktop build. Its source is on GitHub at kwaroran/RisuAI.

No install? The same builds answer in the browser at /chat, with a few free turns a day and no account, and on Telegram.

Where the setting lives in RisuAI

RisuAI ships no model of its own. Every chat goes to whatever you select in its settings, and one entry in that picker is a custom endpoint speaking the OpenAI format. That entry is the one you want.

Its name has moved around. Community write-ups from earlier years call it "Reverse Proxy". In the main branch of RisuAI's source, read on 24 October 2026, the picker entry is Custom API, while the English label on its URL box still reads "Reverse Proxy". Two words, one setting.

Set up the RisuAI custom API in nine steps

  1. Open RisuAI and go to Settings, then Bot Settings.
  2. Change Model to Custom API. Four boxes appear: URL, Key/Password, Request Model and Format.
  3. In URL, paste https://api.unrestrictable.ai/v1/chat/completions.
  4. In Key/Password, paste your key. Ignore the placeholder text about leaving it empty when there is no password: this API wants the key with every call.
  5. In Request Model, enter the model id, such as gemma-4-31b-fabled.
  6. Keep Format at OpenAI Compatible. The remaining choices (Anthropic Claude, Mistral, Google Cloud, OpenAI Response API) build a different request body.
  7. Close settings and open any character.
  8. Send one message.
  9. When a reply arrives, all three of URL, key and model id are correct. If not, the troubleshooting section below sorts the errors by status code.

There is a second way in: RisuAI's Custom Models list, where every entry has its own URL, format and key and an Additional Parameters box. RisuAI's help text says that box adds fields to the request body, and that a key written as header::Name turns into a header. For a single model, the Bot Settings path above is all you need.

The four fields, in one place

Here are the values again as a table, with the labels taken from RisuAI's Bot Settings page and English strings.

URL
https://api.unrestrictable.ai/v1/chat/completions
Whole path, through /chat/completions
Key/Password
sk-unr-live-…
Made in the console; visible once
Request Model
gemma-4-31b-fabled
An id from the Models page, character for character
Format
OpenAI Compatible
Top of the Format list

The reason for pasting the whole path: RisuAI calls the URL exactly as written when it ends in /chat/completions. If Autofill Request URL is switched on, RisuAI will also take https://api.unrestrictable.ai/v1 and add /chat/completions on its own, since that URL ends in v1. The full path is right with the toggle on or off, which is why the guide uses it.

Picking an uncensored build for RisuAI

Each of the four builds here is tagged roleplay in the catalog, and each speaks the OpenAI wire format the Custom API entry produces. Prices and windows below come from the live catalog at render time.

Uncensored models and their prices per million tokens
Wire formats
Gemma 4 31B Fabledgemma-4-31b-fabledRoleplay$0.35$1.575262Kopenai
Qwen3.8 27B Antislopqwen3.8-27b-antislopRoleplay$0.35$2.10262Kopenai
Gemma 4 26B-A4B ChimeraXgemma-4-26b-a4b-chimeraxRoleplay$0.42$1.33262Kopenai

Prices per million tokens, read from the live catalog on 2026-11-16.

Model idCatalog context windowStarting Max Context Size
gemma-4-31b-fabled262,144 tokens32768
qwen3.8-27b-antislop262,144 tokens32768
gemma-4-26b-a4b-chimerax262,144 tokens32768

Windows from the live catalog, 2026-11-16. The starting value is the window, capped at 32,768 tokens.

In RisuAI's strings the context setting is Max Context Size. It governs how much of the card, the lorebook and the chat history travels with each message, and all of it is billed as input tokens. Never set it above the model's window. The Gemma 4 tunes have more relatives on the uncensored Gemma page; the Qwen 3.8 tune has its family on the uncensored Qwen page.

Sampler settings for the custom API entry

In RisuAI's model list the Custom API entry carries a fixed set of samplers: temperature, top_p, frequency_penalty, presence_penalty, repetition_penalty, min_p, top_a, top_k and a budget for thinking tokens. The entry is also flagged for streaming, so text shows up while the model is still writing. Below is the preset we start a roleplay from. Treat every number as a first guess to tune, not as a requirement of the model.

SettingStart valueWhy
Temperature0.9At 1.0 the model samples from its own distribution unchanged. Go down to 0.7 when a build wanders.
Top P0.95Trims the tail of improbable tokens. 1.0 disables it.
Top K0Disabled, since Top P already limits the candidates.
Repetition Penalty1.051.0 means none. A touch above breaks the loops on favourite phrases that long chats fall into.
Frequency and Presence Penalty0Keep both at 0 while Repetition Penalty is on, or the same tokens get punished twice.
Max Response Size300Roughly a paragraph. Builds that reason before answering want 1,000 or more, or the cut lands inside the thinking.

A reply cut off mid-sentence is the Max Response Size limit, not a fault. Raise the value and regenerate. Billing is per token, so a higher ceiling only costs more if the model actually writes more.

The cost of a RisuAI session

Take one sitting of 200 messages with replies near 250 tokens each. In the first column every message carries roughly 4,000 tokens of card and history. In the second it carries 16,000, which is what a lorebook plus a long chat looks like.

Model200 messages at 4,000 context200 messages at 16,000 context
Gemma 4 31B Fabled$0.36$1.20
Qwen3.8 27B Antislop$0.39$1.23
Gemma 4 26B-A4B ChimeraX$0.40$1.41

Worked out at render from catalog prices on 2026-11-16: messages × prompt tokens × input rate, plus messages × 250 × output rate.

Almost all of it is input. 200 × 4,000 comes to 800,000 input tokens, set against 50,000 tokens of output. Trimming the card, pruning the lorebook and lowering Max Context Size change the bill more than swapping models does. The full per-token list is on the pricing page.

Test the key and the id from a terminal

One free request shows every model id your key can call. Put your key in place of the placeholder and run it before opening RisuAI.

curl
1curl https://api.unrestrictable.ai/v1/models \
2 -H "Authorization: Bearer sk-unr-live-…"

A 200 carrying a list of ids confirms the key. When the id you entered in Request Model appears in that list, the URL is the only field that can still be wrong. A 401 points at the key, and the body tells you how:

http
1HTTP 401
2{"error":{"message":"No API key was sent.","type":"authentication_error","param":null,"code":"invalid_api_key"}}
http
1HTTP 401
2{"error":{"message":"The API key is not valid.","type":"authentication_error","param":null,"code":"invalid_api_key"}}

Response bodies captured from the API on 24 October 2026.

RisuAI custom API errors, by status code

401: key

"No API key was sent." says Key/Password is blank. "The API key is not valid." says the key is mistyped, truncated or revoked. Make a new one in the console and paste the whole string.

404: model id, or URL

With the code model_not_found, the Request Model value is not in the catalog: a typo, a display name where an id belongs, or a build that was retired. Rerun the request above and copy the id out of the response. A 404 with no such code means the URL is off, usually a doubled path like /chat/completions/chat/completions or a dropped /v1. Copy the URL from this page once more.

402: balance

Your prepaid credit has run out. Add funds in the console.

CORS in the web build

RisuAI's web build talks to the API directly from the browser. The API accepts requests from every origin, Authorization header included (verified 24 October 2026). Nothing sits in between, and the key travels only to the API.

A fixed prompt goes to each model every day in the daily real-model check, which is how we know the model behind an id matches its label. Base URL, key and id details for other apps are on the docs page.

Questions

The app is open source (kwaroran/RisuAI on GitHub) and free. You pay the API behind the model you connect. Here that is per token, prepaid from $10, with no subscription required. See per-token prices.

It is now the Custom API entry: Settings, Bot Settings, Model. Only the URL box inside it still carries the Reverse Proxy label in RisuAI’s English strings.

https://api.unrestrictable.ai/v1/chat/completions, the whole path. If Autofill Request URL is on, https://api.unrestrictable.ai/v1 works too, since RisuAI adds /chat/completions to any URL ending in v1.

The catalog id, letter for letter, such as gemma-4-31b-fabled. Display names and typos come back as 404 model_not_found. See the uncensored model list.

OpenAI Compatible. The whole catalog speaks the OpenAI wire format, and the other Format choices send a different request body.

No. These builds are abliterated or uncensored fine-tunes: the refusals were removed from the weights, so no prompt text is needed to get past them. 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.

Get a keyPrepaid from $10, no subscription.

New models and price changes, once a month.

New uncensored builds with their prices, price moves, and what the daily model check found.

Join on Telegram