【Open-Source Sharing】Say goodbye to commercial AI censorship and account bans! Deploy any uncensored HuggingFace LLM on a cheap GPU ($0.15/h) with one click using pure Python (supports vLLM/Tabby/llama.cpp)

Technology and Creation AI
1

OneClick Uncensored LLM
OneClick Uncensored LLM1376×768 905 KB

Recently, the content controls and censorship on major commercial LLMs (Claude, GPT, and even some domestic relay services) have been getting increasingly strict, frequently triggering content blocks and even account bans.

Today I’m sharing a handy little tool for tech enthusiasts that was open-sourced recently: OneClick-Uncensored-LLM. Its key features are: built entirely on the Python 3 standard library (zero pip dependencies); it uses an API to automatically rent cost-effective GPUs from Clore.ai with one click (RTX 3090 / 4090 at around $0.08–$0.15/hour), automatically downloads any uncensored model from Hugging Face and deploys it as a standard OpenAI-compatible API, then releases and destroys the instance with one click when testing is finished. The cost is extremely low.

Key features:

  1. Zero-dependency environment: Written entirely with the Python standard library, so it runs directly without the hassle of setting up a complex local Python virtual environment or dealing with pip dependency conflicts.
  2. Fully automatic inference-engine selection based on the model’s weight format:
    • GGUF format → Automatically configures the llama.cpp server (high-speed CUDA acceleration)
    • safetensors / AWQ / GPTQ / FP8 → Automatically configures the vLLM high-concurrency inference engine
    • EXL2 / EXL3 → Automatically configures TabbyAPI (maximizing VRAM usage with ExLlamaV3)
  3. VRAM estimation and cheap-node filtering: Automatically reads the model repository’s config.json to estimate the required VRAM, then hunts for the lowest-priced machine on the market that meets the requirements. If a node fails to respond within 5 minutes or has an excessively slow download speed, the script automatically blacklists it and searches for a better one.
  4. Seamless integration with popular front ends: Once deployment is complete, it automatically outputs a standard /v1/chat/completions API endpoint and a dedicated Bearer Token, which can be entered directly into NextChat, New-API / One-API, Chatbox, or coding plugins such as OpenCode / Cursor / VSCode.
  5. Start and destroy on demand to save money: Once the task or testing is finished, you can instantly release the server with a single click locally, with no unnecessary charges.

Open-source repository:

https://github.com/mooltik/oneclick-uncensored-llm

Recommended uncensored models (available for one-click download with this tool):

  • Qwen2.5-72B-Instruct-Uncensored
  • Llama-3.3-70B-Instruct-Abliterated
  • Mistral-Small-24B-Instruct-2501

Just sharing some tech—if you need to run uncensored models, generate code, or build a low-cost private API node, feel free to give it a try. Welcome to discuss it in the replies!