{"title":"AI Agent Boxes \u0026 Mini PCs","description":"\u003cp\u003eSmall, quiet computers built to run AI around the clock: host always-on agents like OpenClaw or n8n, run open models locally with Ollama and LM Studio, and keep your data on hardware you own.\u003c\/p\u003e\u003ch2\u003eWhat is an AI agent box?\u003c\/h2\u003e\u003cp\u003eAn AI agent box is a mini PC that stays on 24\/7 so your AI assistants can keep working when your laptop is closed. It handles scheduled jobs, inbox triage, content campaigns and home automation, and it can run open-source language models without a monthly cloud bill.\u003c\/p\u003e\u003ch2\u003eHow to choose\u003c\/h2\u003e\u003cul\u003e\n\u003cli\u003e\n\u003cstrong\u003eCloud-connected agents:\u003c\/strong\u003e 32GB of memory is plenty. Start with the GMKtec M5 Ultra.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eAgents plus local models:\u003c\/strong\u003e 64GB and a modern NPU. The EVO-X1 Pro is the sweet spot.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eLarge local LLMs:\u003c\/strong\u003e unified memory matters most. The EVO-X2 with AMD Ryzen AI Max+ 395 is the flagship.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eStorage and AI together:\u003c\/strong\u003e the MINISFORUM N5 Pro is a five-bay AI NAS.\u003c\/li\u003e\n\u003c\/ul\u003e\u003ch2\u003eLocal AI, explained simply\u003c\/h2\u003e\u003cp\u003eA model's size and compression decide how much memory it needs. An 8-billion-parameter model compressed to 4-bit uses about 6GB, a 32B model roughly 20GB, and 70B-class models want 40GB or more. Mini PCs with unified memory let the GPU borrow system RAM, which is why they have become the go-to hardware for running AI at home.\u003c\/p\u003e","products":[{"product_id":"gmktec-evo-x2-ai-mini-pc-ryzen-ai-max-395","title":"GMKtec EVO-X2 AI Mini PC — Ryzen AI Max+ 395, 64GB, 1TB","description":"\u003cp class=\"nx-hook\"\u003eThe mini PC people buy when they want to run serious AI models at home: 16 Zen 5 cores, a 40-unit Radeon GPU and up to 126 TOPS in a box smaller than a hardcover.\u003c\/p\u003e\u003cul class=\"nx-highlights\"\u003e\n\u003cli\u003e\n\u003cstrong\u003eRyzen AI Max+ 395\u003c\/strong\u003e16 cores, 32 threads, boosting to 5.1 GHz.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eRadeon 8060S graphics\u003c\/strong\u003e40 RDNA 3.5 compute units that share the system's fast memory.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eUp to 126 AI TOPS\u003c\/strong\u003eNPU, GPU and CPU combined, per AMD's platform rating.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003e64GB LPDDR5X-8000\u003c\/strong\u003eUnified memory, plus a 1TB PCIe 4.0 SSD and a spare M.2 slot.\u003c\/li\u003e\n\u003c\/ul\u003e\u003ch2\u003eWhy this box is different\u003c\/h2\u003e\u003cp\u003eMost computers split memory between the CPU and a graphics card, and the graphics card is where AI models want to live. That is why a 70-billion-parameter model won't run on a typical 16GB GPU. AMD's Strix Halo chip changes the math: the Radeon 8060S draws from the same pool of LPDDR5X as the processor, so a large share of the machine's memory can go to the model instead.\u003c\/p\u003e\u003cp\u003eIn practice, that means you can load open models through Ollama, LM Studio or llama.cpp that would normally need a workstation GPU, and keep them running around the clock without a cloud bill. Pair it with an agent framework such as OpenClaw, n8n or Home Assistant and it becomes the always-on brain of your home or business.\u003c\/p\u003e\u003ch2\u003eBuilt to stay on\u003c\/h2\u003e\u003cp\u003eA triple-fan cooling design keeps sustained loads in check, and dual M.2 slots let you add storage for models, datasets and backups. Wi-Fi 7, 2.5GbE, USB4 and HDMI 2.1 cover the rest. It drives up to four displays, so it also doubles as a very fast desktop.\u003c\/p\u003e\u003cdiv class=\"nx-callout\"\u003e\u003cp\u003e\u003cstrong\u003eModel sizing tip:\u003c\/strong\u003e 64GB comfortably runs 8B–32B models and quantized 70B-class models with tight context. If you plan to live in 70B territory, ask us about the 128GB configuration.\u003c\/p\u003e\u003c\/div\u003e\u003ch2\u003eSpecifications\u003c\/h2\u003e\u003ctable class=\"nx-specs\"\u003e\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eProcessor\u003c\/td\u003e\n\u003ctd\u003eAMD Ryzen AI Max+ 395 (16C\/32T, up to 5.1 GHz)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eGraphics\u003c\/td\u003e\n\u003ctd\u003eAMD Radeon 8060S, 40 compute units\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eAI engine\u003c\/td\u003e\n\u003ctd\u003eAMD XDNA 2 NPU (up to 50 TOPS); up to 126 TOPS platform\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory\u003c\/td\u003e\n\u003ctd\u003e64GB LPDDR5X-8000 (onboard, unified)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eStorage\u003c\/td\u003e\n\u003ctd\u003e1TB M.2 2280 PCIe 4.0 SSD + second M.2 2280 slot\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eNetworking\u003c\/td\u003e\n\u003ctd\u003eWi-Fi 7, Bluetooth 5.4, 2.5GbE LAN\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003ePorts\u003c\/td\u003e\n\u003ctd\u003eUSB4, USB-A, HDMI 2.1, DisplayPort 1.4, SD card reader, audio\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eDisplays\u003c\/td\u003e\n\u003ctd\u003eUp to 4 simultaneous\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eOperating system\u003c\/td\u003e\n\u003ctd\u003eWindows 11 Pro (Linux friendly)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003ePower plug\u003c\/td\u003e\n\u003ctd\u003eUS\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\u003c\/table\u003e","brand":"GMKtec","offers":[{"title":"Default Title","offer_id":48899321495645,"sku":"INT-GMKEVOX2MIN","price":2199.99,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0743\/9030\/6909\/files\/innaya-hero-35.jpg?v=1791470542"},{"product_id":"gmktec-evo-x1-pro-ai-mini-pc-ryzen-ai-9-hx-470","title":"GMKtec EVO-X1 Pro AI Mini PC — Ryzen AI 9 HX 470, 64GB, 1TB","description":"\u003cp class=\"nx-hook\"\u003eOur pick for most people who want an always-on AI agent that can also run local models: 12 Zen 5 cores, a fast NPU and 64GB of memory in a palm-sized box.\u003c\/p\u003e\u003cul class=\"nx-highlights\"\u003e\n\u003cli\u003e\n\u003cstrong\u003eRyzen AI 9 HX 470\u003c\/strong\u003e12 cores, 24 threads, boosting to about 5.2 GHz.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003e~86 AI TOPS\u003c\/strong\u003eXDNA 2 NPU plus Radeon 890M graphics.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003e64GB LPDDR5X\u003c\/strong\u003eRoom for 8B–32B quantized models alongside your apps.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eDual 2.5GbE + Wi-Fi 7\u003c\/strong\u003eWired and wireless connections for a home server role.\u003c\/li\u003e\n\u003c\/ul\u003e\u003ch2\u003eThe sweet spot for home AI\u003c\/h2\u003e\u003cp\u003eThe EVO-X1 Pro sits between a starter box and a flagship. Its NPU handles background AI tasks at very low power, the Radeon 890M accelerates local inference, and 64GB of memory is enough to keep a capable open model loaded while your agent framework, database and browser automation run next to it.\u003c\/p\u003e\u003cp\u003eThat makes it a natural host for OpenClaw, n8n, Home Assistant or your own Python agents. Leave it on the shelf, connect over the network and let it work while your laptop sleeps.\u003c\/p\u003e\u003ch2\u003eConnected like a server\u003c\/h2\u003e\u003cp\u003eTwo 2.5GbE ports mean you can separate your home network from a lab or NAS network. USB4 and OCuLink leave room to grow, including an external GPU dock later. HDMI, DisplayPort and USB4 video drive three screens when you want to use it as a desktop.\u003c\/p\u003e\u003ch2\u003eSpecifications\u003c\/h2\u003e\u003ctable class=\"nx-specs\"\u003e\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eProcessor\u003c\/td\u003e\n\u003ctd\u003eAMD Ryzen AI 9 HX 470 (12C\/24T)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eGraphics\u003c\/td\u003e\n\u003ctd\u003eAMD Radeon 890M\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eAI engine\u003c\/td\u003e\n\u003ctd\u003eAMD XDNA 2 NPU; about 86 TOPS platform\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory\u003c\/td\u003e\n\u003ctd\u003e64GB LPDDR5X (onboard)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eStorage\u003c\/td\u003e\n\u003ctd\u003e1TB M.2 PCIe 4.0 SSD + second M.2 slot\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eNetworking\u003c\/td\u003e\n\u003ctd\u003eWi-Fi 7, Bluetooth 5.4, dual 2.5GbE LAN\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eExpansion\u003c\/td\u003e\n\u003ctd\u003eUSB4, OCuLink\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eDisplays\u003c\/td\u003e\n\u003ctd\u003eTriple display via HDMI, DisplayPort and USB4\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eOperating system\u003c\/td\u003e\n\u003ctd\u003eWindows 11 Pro\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003ePower plug\u003c\/td\u003e\n\u003ctd\u003eUS\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\u003c\/table\u003e","brand":"GMKtec","offers":[{"title":"Default Title","offer_id":48899321528413,"sku":"INT-GMKEVOX1PRO","price":1599.99,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0743\/9030\/6909\/files\/innaya-hero-34.jpg?v=1791470544"},{"product_id":"gmktec-evo-t1-ai-mini-pc-core-ultra-9-285h","title":"GMKtec EVO-T1 AI Mini PC — Intel Core Ultra 9 285H (Barebone)","description":"\u003cp class=\"nx-hook\"\u003eA tinkerer's AI box: Intel's 16-core Core Ultra 9 285H, Arc 140T graphics and OCuLink for an external GPU. Add your own memory and SSDs and build it exactly the way you want.\u003c\/p\u003e\u003cul class=\"nx-highlights\"\u003e\n\u003cli\u003e\n\u003cstrong\u003eCore Ultra 9 285H\u003c\/strong\u003e16 cores (6P + 8E + 2 LP-E), up to 5.4 GHz.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eUp to 99 platform TOPS\u003c\/strong\u003eNPU, Arc 140T GPU and CPU combined, per Intel.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eDual DDR5-5600 slots\u003c\/strong\u003eInstall up to 128GB of SO-DIMM memory.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eOCuLink + triple M.2\u003c\/strong\u003eBolt on an eGPU and plenty of NVMe storage.\u003c\/li\u003e\n\u003c\/ul\u003e\u003ch2\u003eBarebone means your rules\u003c\/h2\u003e\u003cp\u003eThis version ships without RAM, storage or an operating system. That is deliberate: it lets you choose how much memory your models need, which SSDs you trust and whether you run Windows, Ubuntu or Proxmox. Two SO-DIMM slots take up to 128GB of DDR5, and three M.2 slots give you room for a boot drive, a model library and backups.\u003c\/p\u003e\u003ch2\u003eRoom to grow into GPU inference\u003c\/h2\u003e\u003cp\u003eThe OCuLink port connects straight to PCIe, so pairing the EVO-T1 with an eGPU dock and a desktop graphics card turns it into a compact local-LLM rig without the bandwidth penalty of Thunderbolt. Dual 2.5GbE ports make it a strong homelab or virtualization host.\u003c\/p\u003e\u003cdiv class=\"nx-callout\"\u003e\u003cp\u003e\u003cstrong\u003eGood to know:\u003c\/strong\u003e Memory, SSD and operating system are not included. You'll need DDR5 SO-DIMM modules and at least one M.2 2280 NVMe SSD to get started.\u003c\/p\u003e\u003c\/div\u003e\u003ch2\u003eSpecifications\u003c\/h2\u003e\u003ctable class=\"nx-specs\"\u003e\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eProcessor\u003c\/td\u003e\n\u003ctd\u003eIntel Core Ultra 9 285H (16 cores, up to 5.4 GHz)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eGraphics\u003c\/td\u003e\n\u003ctd\u003eIntel Arc 140T\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eAI engine\u003c\/td\u003e\n\u003ctd\u003eIntel AI Boost NPU; up to 99 TOPS platform\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory\u003c\/td\u003e\n\u003ctd\u003e2 x DDR5-5600 SO-DIMM, up to 128GB (not included)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eStorage\u003c\/td\u003e\n\u003ctd\u003e3 x M.2 2280 PCIe 4.0 slots (SSD not included)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eNetworking\u003c\/td\u003e\n\u003ctd\u003eWi-Fi 6, Bluetooth 5.2, dual 2.5GbE LAN\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eExpansion\u003c\/td\u003e\n\u003ctd\u003eOCuLink, USB4\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eOperating system\u003c\/td\u003e\n\u003ctd\u003eNone (barebone)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003ePower plug\u003c\/td\u003e\n\u003ctd\u003eUS\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\u003c\/table\u003e","brand":"GMKtec","offers":[{"title":"Default Title","offer_id":48899321593949,"sku":"INT-GMKEVOT1MIN","price":829.99,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0743\/9030\/6909\/files\/innaya-hero-33.jpg?v=1791470543"},{"product_id":"gmktec-m5-ultra-mini-pc-ai-agent-starter","title":"GMKtec M5 Ultra Mini PC — Ryzen 7 7730U, 32GB, 1TB (AI Agent Starter Box)","description":"\u003cp class=\"nx-hook\"\u003eThe simplest way to give your AI agent a permanent home. Eight Ryzen cores, 32GB of memory and a 1TB SSD in a quiet box that sips power and never needs to be closed like a laptop.\u003c\/p\u003e\u003cul class=\"nx-highlights\"\u003e\n\u003cli\u003e\n\u003cstrong\u003eRyzen 7 7730U\u003c\/strong\u003e8 cores, 16 threads, up to 4.5 GHz.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003e32GB DDR4 + 1TB SSD\u003c\/strong\u003eUpgradable to 64GB, with a second M.2 slot.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eDual 2.5GbE LAN\u003c\/strong\u003eServer-style networking out of the box.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eLow power, low noise\u003c\/strong\u003eBuilt for 24\/7 desk or shelf duty.\u003c\/li\u003e\n\u003c\/ul\u003e\u003ch2\u003eWho this is for\u003c\/h2\u003e\u003cp\u003eMost AI agents do their heavy thinking in the cloud. OpenClaw, n8n and similar tools call models like Claude or GPT, then act on the results: sorting email, posting content, updating spreadsheets, watching for events. What they need locally is a reliable computer that stays on. That is exactly what the M5 Ultra is.\u003c\/p\u003e\u003cp\u003eIt will also run small open models locally through Ollama (think 3B–8B for quick classification or summaries), but it has no dedicated NPU, so if local AI is your main goal, step up to the EVO-X1 Pro.\u003c\/p\u003e\u003ch2\u003eSpecifications\u003c\/h2\u003e\u003ctable class=\"nx-specs\"\u003e\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eProcessor\u003c\/td\u003e\n\u003ctd\u003eAMD Ryzen 7 7730U (8C\/16T, up to 4.5 GHz)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eGraphics\u003c\/td\u003e\n\u003ctd\u003eAMD Radeon integrated graphics\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory\u003c\/td\u003e\n\u003ctd\u003e32GB DDR4-3200 dual channel (up to 64GB)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eStorage\u003c\/td\u003e\n\u003ctd\u003e1TB M.2 NVMe SSD + second M.2 slot\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eNetworking\u003c\/td\u003e\n\u003ctd\u003eDual 2.5GbE LAN, Wi-Fi 6E, Bluetooth 5.2\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eDisplays\u003c\/td\u003e\n\u003ctd\u003eTriple display output\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eOperating system\u003c\/td\u003e\n\u003ctd\u003eWindows 11 Pro\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003ePower plug\u003c\/td\u003e\n\u003ctd\u003eUS\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\u003c\/table\u003e","brand":"GMKtec","offers":[{"title":"Default Title","offer_id":48899321626717,"sku":"INT-GMKM5ULTMIN","price":599.99,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0743\/9030\/6909\/files\/innaya-hero-32.jpg?v=1791470544"},{"product_id":"minisforum-n5-pro-ai-nas","title":"MINISFORUM N5 Pro AI NAS — Ryzen AI 9 HX PRO 370, 5-Bay, 10GbE","description":"\u003cp class=\"nx-hook\"\u003eYour private cloud and your AI server in one chassis. Five drive bays, 10GbE networking and a Ryzen AI 9 HX PRO 370 that can index, search and summarize everything you store.\u003c\/p\u003e\u003cul class=\"nx-highlights\"\u003e\n\u003cli\u003e\n\u003cstrong\u003eRyzen AI 9 HX PRO 370\u003c\/strong\u003e12 cores, 24 threads, up to 5.1 GHz.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003e80 AI TOPS\u003c\/strong\u003eXDNA 2 NPU plus Radeon 890M graphics.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003e5 bays + M.2\u003c\/strong\u003eUp to 144TB of raw capacity with large drives.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003e10GbE + 5GbE\u003c\/strong\u003eFast enough to edit video straight off the NAS.\u003c\/li\u003e\n\u003c\/ul\u003e\u003ch2\u003eStorage that thinks\u003c\/h2\u003e\u003cp\u003eTraditional NAS boxes use low-power chips that struggle with anything beyond file serving. The N5 Pro uses a laptop-class Ryzen AI processor, so it can run photo recognition, local search, media transcoding and small language models right next to your data. Nothing has to leave your house to be understood.\u003c\/p\u003e\u003cp\u003eIt also makes an excellent home for AI agents and self-hosted apps. Run TrueNAS, Unraid or Proxmox, spin up containers for your automations and keep a local model ready for private questions about your own files.\u003c\/p\u003e\u003cdiv class=\"nx-callout\"\u003e\u003cp\u003e\u003cstrong\u003eConfiguration:\u003c\/strong\u003e 16GB DDR5 with a 128GB OS drive. Hard drives and data SSDs are not included. ECC memory is supported if you want to upgrade.\u003c\/p\u003e\u003c\/div\u003e\u003ch2\u003eSpecifications\u003c\/h2\u003e\u003ctable class=\"nx-specs\"\u003e\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eProcessor\u003c\/td\u003e\n\u003ctd\u003eAMD Ryzen AI 9 HX PRO 370 (12C\/24T)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eAI engine\u003c\/td\u003e\n\u003ctd\u003eXDNA 2 NPU; up to 80 TOPS platform\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eGraphics\u003c\/td\u003e\n\u003ctd\u003eAMD Radeon 890M\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory\u003c\/td\u003e\n\u003ctd\u003e16GB DDR5 (ECC supported, upgradable)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eOS drive\u003c\/td\u003e\n\u003ctd\u003e128GB SSD included\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eDrive bays\u003c\/td\u003e\n\u003ctd\u003e5 x SATA bays + M.2 NVMe slots (drives not included)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eNetworking\u003c\/td\u003e\n\u003ctd\u003e10GbE + 5GbE LAN\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eExpansion\u003c\/td\u003e\n\u003ctd\u003eOCuLink, USB4, PCIe slot\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003ePower plug\u003c\/td\u003e\n\u003ctd\u003eUS\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\u003c\/table\u003e","brand":"MINISFORUM","offers":[{"title":"Default Title","offer_id":48899321659485,"sku":"INT-MINN5PRONAS","price":1179.0,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0743\/9030\/6909\/files\/innaya-hero-31.jpg?v=1791470544"},{"product_id":"minisforum-m2-pro-ai-mini-pc","title":"MINISFORUM M2 Pro AI Mini PC — Intel Core Ultra X7 358H, 32GB, 1TB","description":"\u003cp class=\"nx-hook\"\u003eIntel's newest AI silicon in a 1.8-litre box: a 16-core Core Ultra X7 358H and an Arc B390 GPU rated at 122 TOPS, plus 10-gigabit Ethernet and three USB4 ports.\u003c\/p\u003e\u003cul class=\"nx-highlights\"\u003e\n\u003cli\u003e\n\u003cstrong\u003eCore Ultra X7 358H\u003c\/strong\u003e16 cores, up to 4.8 GHz.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eArc B390 GPU\u003c\/strong\u003e12 Xe3 cores rated at 122 TOPS INT8, plus a 50 TOPS NPU.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003e10GbE + 2.5GbE\u003c\/strong\u003eCopper 10-gig on RJ45, no transceiver needed.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003e3 x M.2 slots\u003c\/strong\u003eUp to 24TB of NVMe storage.\u003c\/li\u003e\n\u003c\/ul\u003e\u003ch2\u003eA GPU that punches above its size\u003c\/h2\u003e\u003cp\u003eThe M2 Pro's headline is its graphics. The Arc B390 moves most AI inference onto the GPU, which is where llama.cpp and Stable Diffusion backends want to run, while the NPU handles low-power background jobs. The result is a compact machine that is equally comfortable as an AI workstation, a home server or a quiet 4K desktop.\u003c\/p\u003e\u003cp\u003eMemory is 32GB of onboard LPDDR5X at up to 7467 MT\/s. It is fast, but it is soldered, so choose this box for agents, coding assistants and 8B–14B models rather than giant local LLMs.\u003c\/p\u003e\u003ch2\u003eServer-grade connectivity\u003c\/h2\u003e\u003cp\u003e10GbE on RJ45 sits next to 2.5GbE, Wi-Fi 7 and Bluetooth 6.0. Three USB4 ports, HDMI 2.1 (4K at 144Hz or 8K at 60Hz) and DisplayPort 1.4 drive up to four screens. The power supply is built in, so there is no brick on the floor.\u003c\/p\u003e\u003ch2\u003eSpecifications\u003c\/h2\u003e\u003ctable class=\"nx-specs\"\u003e\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eProcessor\u003c\/td\u003e\n\u003ctd\u003eIntel Core Ultra X7 358H (16 cores, up to 4.8 GHz)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eGraphics\u003c\/td\u003e\n\u003ctd\u003eIntel Arc B390, 12 Xe3 cores (122 TOPS INT8)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eNPU\u003c\/td\u003e\n\u003ctd\u003e50 TOPS INT8\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory\u003c\/td\u003e\n\u003ctd\u003e32GB LPDDR5X, up to 7467 MT\/s (onboard)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eStorage\u003c\/td\u003e\n\u003ctd\u003e1TB SSD; 3 x M.2 2280 slots (2 x PCIe 4.0 x4, 1 x PCIe 4.0 x1), up to 24TB\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eNetworking\u003c\/td\u003e\n\u003ctd\u003e10GbE RJ45 + 2.5GbE RJ45, Wi-Fi 7, Bluetooth 6.0\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003ePorts\u003c\/td\u003e\n\u003ctd\u003e3 x USB4, HDMI 2.1, DisplayPort 1.4\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003ePower\u003c\/td\u003e\n\u003ctd\u003eBuilt-in 134.9W power supply\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eSize and weight\u003c\/td\u003e\n\u003ctd\u003e195 x 195 x 47.5 mm, 1.39 kg\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\u003c\/table\u003e","brand":"MINISFORUM","offers":[{"title":"Default Title","offer_id":48899321692253,"sku":"INT-MINM2PROMIN","price":1499.0,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0743\/9030\/6909\/files\/innaya-hero-30.jpg?v=1791470543"},{"product_id":"minisforum-deg1-oculink-egpu-dock","title":"MINISFORUM DEG1 OCuLink eGPU Dock — PCIe 4.0 x4 External GPU Station","description":"\u003cp class=\"nx-hook\"\u003eThe cheapest way to turn a mini PC into a serious local-AI machine: plug a full-size desktop graphics card into your OCuLink port.\u003c\/p\u003e\u003cul class=\"nx-highlights\"\u003e\n\u003cli\u003e\n\u003cstrong\u003eOCuLink PCIe 4.0 x4\u003c\/strong\u003eUp to 64Gbps straight to the CPU.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eFull-size GPU support\u003c\/strong\u003eOpen-frame design fits large triple-fan cards.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eATX or SFX power\u003c\/strong\u003eUse the PSU you already own.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eAluminum frame\u003c\/strong\u003eSturdy, compact and easy to cool.\u003c\/li\u003e\n\u003c\/ul\u003e\u003ch2\u003eWhy OCuLink beats Thunderbolt for AI\u003c\/h2\u003e\u003cp\u003eOCuLink exposes raw PCIe lanes, so there is no controller in the middle and far less overhead than a Thunderbolt enclosure. For local LLMs that means your graphics card's VRAM and compute are available at close to desktop speed. It is a popular upgrade for GMKtec and MINISFORUM boxes with an OCuLink port.\u003c\/p\u003e\u003cdiv class=\"nx-callout\"\u003e\u003cp\u003e\u003cstrong\u003eYou'll need:\u003c\/strong\u003e a mini PC or laptop with an OCuLink port (or an M.2-to-OCuLink adapter), a desktop graphics card and an ATX or SFX power supply. GPU and PSU are not included.\u003c\/p\u003e\u003c\/div\u003e\u003ch2\u003eSpecifications\u003c\/h2\u003e\u003ctable class=\"nx-specs\"\u003e\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eInterface\u003c\/td\u003e\n\u003ctd\u003eOCuLink (PCIe 4.0 x4), up to 64Gbps\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eGPU slot\u003c\/td\u003e\n\u003ctd\u003ePCIe x16 physical\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003ePower\u003c\/td\u003e\n\u003ctd\u003eATX or SFX power supply (not included)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMaterial\u003c\/td\u003e\n\u003ctd\u003eAluminum alloy, open frame\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCompatibility\u003c\/td\u003e\n\u003ctd\u003eDevices with OCuLink 4i port\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\u003c\/table\u003e","brand":"MINISFORUM","offers":[{"title":"Default Title","offer_id":48899321725021,"sku":"INT-MINDEGOCUEGP","price":129.99,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0743\/9030\/6909\/files\/innaya-hero-29.jpg?v=1791470543"},{"product_id":"m2-nvme-to-oculink-adapter-sff-8612","title":"M.2 NVMe to OCuLink Adapter — M-Key 2230–2280 to SFF-8612, PCIe 4.0 x4","description":"\u003cp class=\"nx-hook\"\u003eTurn a spare M.2 slot into an OCuLink port. This adapter card brings PCIe x4 out of your mini PC or desktop through a standard SFF-8612 connector.\u003c\/p\u003e\u003cul class=\"nx-highlights\"\u003e\n\u003cli\u003e\n\u003cstrong\u003eM.2 M-key\u003c\/strong\u003eFits 2230, 2242, 2260 and 2280 slots.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eOCuLink output\u003c\/strong\u003eSFF-8612 female connector.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003ePCIe 3.0 \/ 4.0\u003c\/strong\u003eFour-lane (x4) link.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eCompact card\u003c\/strong\u003eMounts like an NVMe SSD.\u003c\/li\u003e\n\u003c\/ul\u003e\u003ch2\u003eExpand over OCuLink\u003c\/h2\u003e\u003cp\u003eUse an OCuLink cable to connect U.2\/U.3 SSDs or other OCuLink devices, such as an external GPU dock for local AI and homelab builds.\u003c\/p\u003e\u003cp\u003e\u003cem\u003eAdapter card only. OCuLink cable and devices not included. Check that your M.2 slot supports PCIe NVMe.\u003c\/em\u003e\u003c\/p\u003e\u003ch2\u003eSpecifications\u003c\/h2\u003e\u003ctable class=\"nx-specs\"\u003e\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eModel\u003c\/td\u003e\n\u003ctd\u003eAC8019\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eInput\u003c\/td\u003e\n\u003ctd\u003eM.2 M-key, PCIe x4\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eLengths\u003c\/td\u003e\n\u003ctd\u003e2230, 2242, 2260, 2280\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eOutput\u003c\/td\u003e\n\u003ctd\u003eOCuLink SFF-8612 (female)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003ePCIe support\u003c\/td\u003e\n\u003ctd\u003ePCIe 3.0 and 4.0\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eUse with\u003c\/td\u003e\n\u003ctd\u003eU.2\/U.3 SSDs, OCuLink devices\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\u003c\/table\u003e","brand":"Innaya Tech","offers":[{"title":"Adapter card","offer_id":67787475746909,"sku":"INT-M2OCU-8612","price":21.99,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0743\/9030\/6909\/files\/innaya-hero-m2-oculink-adapter.jpg?v=1792425418"}],"url":"https:\/\/innaya.tech\/collections\/ai-agent-boxes.oembed","provider":"Innaya Tech","version":"1.0","type":"link"}