This dude casually runs Qwen3.5-27B fully local on Telegram.

No APIs, no costs, no data sent anywhere, all sandboxed with NVIDIA NemoClaw on a DGX Spark.

9 t/s and smooth chats. The future might be local indeed.