A small plug-and-play hub that connects the computers, Macs, and phones you already own into one pooled-memory network — so you can run AI models too big for any single device, without the cloud.
RAMDeck turns the computers, phones and other devices you already own into a single, unified local AI cluster by pooling their RAM. A small hub acts as a conductor, directing model layers to each device so you can run large language models locally without recurring cloud fees or buying expensive new hardware. The kit includes a pre‑flashed hub, node agents for Windows/macOS/Android, an Ethernet cable, and OpenAI‑compatible API access, enabling self‑healing, dynamic context scaling and local knowledge‑base ingestion. With shipping estimated for February‑March 2027, RAMDeck offers a one‑time cost against the endless expense of cloud subscriptions.
| Spec | Details |
|---|---|
| Power Adapter | 5V/4A (US only) |
| Ethernet Cable | Included |
| Supported OS | Windows, macOS, Android |
| API Compatibility | OpenAI‑compatible (/v1/chat/completions) |
| Shipping Estimate | February – March 2027 |
| Retail Price | $199 |
| Source Availability | GitHub (Apache 2.0 with Commons Clause) |
| Dynamic Context Scaling | Up to 32,000+ tokens with pooled RAM |
Performance depends on network stability; Wi‑Fi drops or sleeping nodes can interrupt inference and require model reloads, which take minutes for large models. Wired Ethernet and keeping devices awake are recommended to minimize disruptions.
Development progress, shipping timeline, and announcements
View all updates ↗