Summary
This video walks through how to run an uncensored AI model without the refusals found in mainstream chatbots. Creator Jack Roberts demonstrates an “obliterated” version of DeepSeek V4.1 Flash, connected through obliteration.ai, and shows how it can be used inside the Claude Code harness to keep the tooling and performance of a modern coding agent while removing content filters.
The tutorial is organized into five levels. It starts with generating an API key and plugging the model into a custom Claude-style interface that supports chats, artifacts, spreadsheets, documents, slide decks, whiteboards and dashboards, with the option to swap between models such as Qwen, Claude Code and Codex. Later sections cover running the setup on a rented virtual private server, connecting devices with Tailscale, and reaching it from a laptop or phone.
Roberts also makes the case for privacy, noting that chats with hosted AI services can be accessed by the provider, by courts, and potentially by governments. He explains the difference between a model and a harness, and offers a guide of copyable prompts. Viewers interested in open-weight models, self-hosting and private AI workflows will find a practical starting point, along with a stated warning to use these tools responsibly.
📺 Source: Jack Roberts · Published November 15, 2026
🏷️ Format: Tutorial Demo







