$109 RAMDeck Hub Pools Memory Across Devices to Run Larger AI Models

TradeMAV is crowdfunding RAMDeck, a $109 hub designed to connect Windows PCs, Macs, Linux machines and Android phones on a home network. It distributes an AI model across the devices so it can run even when the model does not fit in any one machine’s memory. The campaign has a $10,000 funding goal and is scheduled to close on November 8, 2026.

RAMDeck does not combine memory at the hardware level. Instead, each device holds part of the model, and intermediate results move across the network as it generates tokens. TradeMAV says its software handles device discovery, assigns model layers, recovers from failures and provides a dashboard and an OpenAI-compatible API. Its node agent uses the existing llama.cpp RPC server; the company’s guide says users comfortable with terminal scripts can use a free, open-source DIY approach.

The hub’s processor, memory, ports and power draw are not listed in TradeMAV’s published materials. The company also warns that pooled memory trades speed for capacity. Its benchmarks include 39 tokens per second for Qwen3-Coder 30B across a Beelink mini PC and an M4 Mac mini, and 1.37 tokens per second for gpt-oss-120b on a six-device cluster. TradeMAV says the hub offers convenience, while its underlying software is available for free.