Uncensored AI: where to use it, how much it costs, and what to check

Call Venice through its OpenAI-compatible API

Venice follows the OpenAI specification, so any OpenAI library works if you change the endpoint. Create a key at venice.ai/settings/api and store it in your terminal. The API uses credits: the subscription includes only a small monthly allowance (Pro: US$ 1; free accounts: none), so add funds first. Without a balance, the API returns 402 INSUFFICIENT_BALANCE.

Linux, Mac, or WSL (bash terminal):

export VENICE_API_KEY="paste-your-key-here"   # never put it in code or git

curl https://api.venice.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"venice-uncensored-1-2","messages":[{"role":"user","content":"Explain what abliteration is in 3 sentences."}]}'

Windows (PowerShell):

$env:VENICE_API_KEY="paste-your-key-here"   # valid only for this window

$body = @{ model="venice-uncensored-1-2"; messages=@(@{ role="user"; content="Explain what abliteration is in 3 sentences." }) } | ConvertTo-Json -Depth 5
Invoke-RestMethod -Uri https://api.venice.ai/api/v1/chat/completions -Method Post -Headers @{ Authorization="Bearer $env:VENICE_API_KEY" } -ContentType "application/json; charset=utf-8" -Body $body

The same in Python, using the openai library. First create an isolated environment so you do not change the system Python:

# Linux, Mac, or WSL
python3 -m venv .venv && source .venv/bin/activate
pip install openai

# Windows (PowerShell)
py -m venv .venv; .venv\Scripts\Activate.ps1
pip install openai

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.venice.ai/api/v1",
    api_key=os.environ["VENICE_API_KEY"],
)
r = client.chat.completions.create(
    model="deepseek-v4-1-flash-1017-uncensored",  # or venice-uncensored-1-2, gemma-4-uncensored
    messages=[{"role": "user", "content": "List 3 risks of an uncensored model."}],
)
print(r.choices[0].message.content)

Some models are available only on paid plans and return a 401 PRO_ONLY_MODEL error. IDs change: check the current list at docs.venice.ai. abliteration.ai works the same way, using base_url="https://api.abliteration.ai/v1" and its own key.

Run a model on your computer with Ollama

First install Ollama. On Windows and Mac, download it from ollama.com/download. On Linux:

curl -fsSL https://ollama.com/install.sh | sh

Memory rule: the model must fit in your graphics card’s memory (VRAM), with 1 to 2 GB to spare for the conversation. The Ollama file listing shows its size: orcarouter/Qwen3.8-27B-Uncensored (the default quantized version) is 18 GB, so it needs a 24 GB GPU. Without a GPU it runs in your computer’s memory, but much more slowly.

Choose based on your machine (sizes checked in the Ollama registry on Nov 9, 2026):

# GPU with 24 GB or more (18 GB file)
ollama run orcarouter/Qwen3.8-27B-Uncensored

# 8 GB GPU (5 GB file)
ollama run huihui_ai/qwen3-abliterated:8b

# CPU only or basic laptop (2.5 GB file)
ollama run huihui_ai/qwen3-abliterated:4b

The first time, the command downloads the model; after that, it opens the chat directly. To exit, type /bye.

# the same model through Ollama’s local API (Ollama must be running;
# if you get a connection error, run "ollama serve" in another terminal)
curl http://localhost:11434/api/chat -d '{"model":"orcarouter/Qwen3.8-27B-Uncensored","messages":[{"role":"user","content":"hi"}],"stream":false}'

Want to set this up for access from your phone without exposing anything to the internet, and compare a standard model with an uncensored one? The related guide walks you through the whole process: Private AI at 3 levels.

Connect the model to an agent tool

Agent tools and editors that accept an “OpenAI-compatible endpoint” can use any of these models. The idea is always the same: three fields:

• Base URL: https://api.venice.ai/api/v1, https://api.abliteration.ai/v1, or locally, http://localhost:11434/v1.
• Key: your own, stored in an environment variable (with local Ollama, the field is required but ignored; its documentation uses ollama).
• Model name: the exact ID from the table, such as venice-uncensored-1-2.

Concrete example: opencode accepts a custom provider in the opencode.jsonc file, using the format in its documentation. For Venice:

{
  "$schema": "https://opencode.ai/config.json",
  "providers": {
    "venice": {
      "name": "Venice",
      "env": ["VENICE_API_KEY"],
      "package": "@opencode/ai/providers/openai-compatible",
      "settings": { "baseURL": "https://api.venice.ai/api/v1" },
      "models": { "venice-uncensored-1-2": {} }
    }
  }
}

The model appears as venice/venice-uncensored-1-2. With local Ollama, opencode finds models automatically at http://127.0.0.1:11434. Other agent tools, such as Claude Code and Codex, are designed for their companies’ own models and may limit or reject outside providers; check each tool’s documentation before trying. abliteration.ai also supports Anthropic’s format, which helps with tools that accept only that format. Remember: a smaller model without guardrails makes more mistakes on long agent tasks, so review every step it takes.