Who We Are and Why We Built This
We’ve successfully funded, built, and shipped complex hardware and software products before. RAMDeck is built on that same foundation of delivery.
The RAMDeck Origin Story
RAMDeck exists because we refused to pay exorbitant cloud fees or buy overpriced hardware just to scale our AI.
We run TradeMAV, a highly successful, local AI-powered market research and technical analysis tool that runs entirely from a USB drive. With a rapidly expanding global user base that demands premium technical support, we needed to scale our infrastructure efficiently. We built a powerful local, RAG-based AI assistant — we call it Jarvis — trained directly on our proprietary knowledge base. Initially running on a single Mac Mini, Jarvis handled the complex, high-volume inquiries so our human experts could focus on edge-cases.
As our knowledge base and our models grew, one Mac Mini stopped being enough. And this happened right as the industry hit what’s now being called the “RAMpocalypse” — a real, ongoing global memory shortage. AI datacenter demand for HBM memory has been crowding out consumer DRAM production since 2025, and the price swings have been brutal: a 64GB DDR5 kit that cost around $191 in August 2025 was going for $1,118 a year later, and industry analysts don’t expect real relief until 2027 or later. Buying our way out of the problem with more RAM simply wasn’t an option.
So we did the only thing that made sense: we pointed every device we already owned — old PCs, a gaming rig, a Mac, whatever had spare RAM or VRAM — at the same problem, and taught them to work together as one pool of memory instead of buying a bigger single machine.
That pooled-hardware approach is RAMDeck. We still run models like Qwen across our own cluster today. This page exists because it worked well enough that we think you should have one too.
⚠️ IMPORTANT DISCLAIMER
RAMDeck is a local inference orchestration tool. It does NOT guarantee specific inference speeds, token-per-second rates, or hardware miracles.
Performance depends on your hardware: Your results depend entirely on the specific computers you connect, your local network speed, and the model you choose to run.
Pooled memory trades speed for capacity: Distributing a model across a network is slower than running it entirely on a dedicated $10,000 GPU. It is designed to let you run models that would not fit otherwise, not to beat dedicated data-center hardware.
We provide the engine and a working dashboard, you provide the rest. For building custom applications, you’ll connect your own frontend — Continue.dev, LangChain, or a custom script — via the OpenAI-compatible API endpoint.
We are not responsible for your data. While RAMDeck keeps all data locally on your LAN by default, you are solely responsible for securing your own local network against internal threats or misconfigured firewall exposures.
Take Back Control of Your AI
Our pre-launch campaign is now live. Join the waitlist on our Indiegogo page to get priority access when we launch.
View Indiegogo Campaign