DEV Community

Tommy T
Tommy T

Posted on Edited on

Using an uncensored LLM through an OpenAI-compatible API (Python and CLI)

Most hosted LLMs refuse a lot of requests that are perfectly legitimate: security research, fiction with dark themes, medical or legal questions. I built Orrevan to have an alternative: uncensored (abliterated) open models behind a normal OpenAI-compatible API, plus a terminal CLI.

This post shows how to use it in a few lines. Disclosure: I am the developer of Orrevan.

What it is

  • Chat, API and CLI for uncensored and abliterated models (DeepSeek, GLM, Qwen and MiMo variants)
  • Prepaid credit, billed per token
  • Chats are encrypted with AES-256-GCM before they are stored, no training on your data
  • OpenAI-compatible, so existing SDKs work by changing the base URL

Quick start with curl

curl https://orrevan.com/api/v1/chat/completions \
  -H "Authorization: Bearer $ORREVAN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "orrevan/abliterated-model",
       "messages": [{"role": "user", "content": "Hello"}]}'
Enter fullscreen mode Exit fullscreen mode

Python with the openai package

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://orrevan.com/api/v1",
    api_key=os.environ["ORREVAN_API_KEY"],
)
r = client.chat.completions.create(
    model="orrevan/abliterated-model",
    messages=[{"role": "user", "content": "Hello"}],
)
print(r.choices[0].message.content)
Enter fullscreen mode Exit fullscreen mode

Keep the key on the server. Never put it into browser code or a public repo.

The CLI

You need Node.js 20 or newer.

npm i -g orrevan
orrevan login          # save your key
orrevan                # open a session in the current folder
orrevan -p "question"  # answer once and exit
orrevan models         # list models
orrevan balance        # check your credit
Enter fullscreen mode Exit fullscreen mode

Models

There are 10 models in total, for example orrevan/deepseek-v4.1-flash-unc, orrevan/glm53-flash-unc, orrevan/qwen38-unc and orrevan/mimo-v26-flash-unc. One model has a 1M token context window and some understand images. Streaming is supported.

Feedback

If you try it, I would like to hear what is missing or broken. You can reach me via the site: https://orrevan.com

Top comments (0)