WireGuard + DGX Spark: give any device secure access to 128GB of local LLM memory
aestheticedwar1 · x · 2026-10-14
A home networking setup for local LLMs: WireGuard connects a phone (Ollama app), laptop (editor/scripts), and home server (n8n, agents) to a DGX Spark with 128GB unified memory running gpt-oss:20b via Ollama on port 11434.
- No port forwards on the router; access is limited to the WireGuard tunnel and a TCP 11434 policy
- Public scanners find nothing to reach — no firewall holes or exposed SSH needed
- Any device can securely use the home model box, effectively giving your laptop 128GB of memory it doesn't have
More from coding & agent
- Orange Built a 120,000-Employee AI Platform on LangChain and LangSmith — LangChain · 2026-10-14
- Every Open-Sources Compound Writing, a Context-First AI Writing Toolbox for Claude and Codex — every · 2026-10-14
- Cut capture friction: one button per section in Obsidian daily notes — dSebastien · 2026-10-14
- AI agent rule goes viral: even "hand jobs" must route through the agent-scratch runner — StewartalsopIII · 2026-10-14
- Job queue stuck for two days: the case for leases over a two-state pending/done flag — Kodenstudio · 2026-10-14
- Cohere Is Hiring for Its RL Environments Team Building Enterprise Replicas to Train AI Agents — hugo_larochelle · 2026-10-14