Wolfgang’s Channel

YouTube creator focused on homelab and self-hosting, with strong opinions about running services on your own hardware rather than renting them. Editorial position is privacy-and-control-first, similar in spirit to Crosstalk Solutions and Joshua Clarke but with a sharper “this is a homelab channel, you’ve been looking for stuff to self-host anyway” framing.

Channels

• YouTube: Wolfgang’s Channel — homelab, self-hosting, Docker, Linux, hardware reviews

Content in This Wiki

• Host Your Own AI Code Assistant with Docker, Ollama and Continue! — Real-world comparison of self-hosting a local Copilot replacement on a 7900XTX gaming PC vs a LattePanda Sigma MiniPC. Hard data on power consumption (130W vs 4.6W idle), and the practical conclusion that without a real GPU, local code assistance is unusable. Introduces the Continue VS Code extension as the local-Copilot frontend and walks through the Docker Compose recipe with ROCm device passthrough for AMD GPUs.

Key Ideas

• GitHub Copilot leaked API keys → self-hosting became a requirement for code assistance, not just a preference. The privacy threat is concrete, not abstract.
• “What I want from an AI code assistant is more intelligent, context-aware auto-suggestions, not whole-code generation” — Wolfgang’s framing is anti-vibe-coding: he wants the machine to suggest the boilerplate it knows he’s about to type, not write the program for him. This is closer to JetBrains’ historical line completion than to Cursor’s compose.
• The GPU vs MiniPC gap is a cliff, not a slope — at 130W average draw the gaming rig produces useful suggestions, at 4.6W idle the MiniPC produces unusable hallucinations. There is no middle ground for the workloads he tested. Aligns with Alex Ziskind’s vLLM analysis about why parallelism matters even for solo developers.
• AMD ROCm is fully supported by Ollama on Ubuntu (not Debian — community consensus is to use Ubuntu for ROCm). The 7900XTX is a viable alternative to NVIDIA for local AI if you’re willing to debug ROCm setup.
• “Run Ollama on a separate device so you can share it with a friend or use it on multiple devices” — homelab framing, treats local AI as shared infrastructure, not per-device

See Also

• continue — the VS Code extension Wolfgang uses
• ollama — the local LLM runtime
• open-webui — the chat UI Wolfgang pairs with Ollama in Docker Compose
• crosstalk-solutions, joshua-clarke — adjacent self-hosting / homelab voices in the wiki
• alex-ziskind — adjacent local-AI hardware deep-dives (Wolfgang’s power-vs-throughput conclusions corroborate Alex’s vLLM benchmarks)
• summary-wolfgangs-channel-self-host-ai-code-assistant