RAMDeck v0.1.0 - Distributed Local Inference Engine

RAMDeck v0.1.0: First Source-Available Release

RAMDeck networks heterogeneous consumer devices over local Wi-Fi or Gigabit Ethernet, partitions an LLM's transformer layers across each computer's physical memory, and orchestrates them into a unified local AI server.

Features

• mDNS Auto-Discovery: Nodes discover each other on the local network automatically.
• Dynamic Memory Waterfill: Packs fast GPUs first, then host RAM, then remote network peers.
• Heterogeneous Cluster Support: Run across Windows (CUDA/CPU), macOS (Metal), Linux, and Android.
• Self-Healing Architecture: Recovers from dropped nodes and closed laptop lids by redistributing shards.
• OpenAI-Compatible API: Exposes localhost:8420/v1/chat/completions for Open-WebUI, Cursor, and LibreChat.

Full documentation & benchmarks: https://ramdeck.com/how-to-pool-ram-from-old-pcs/