AI mini PCs
Compact boxes with enough RAM for 7B–14B local models and always-on agent workers. Look for 32GB+ unified or discrete memory.
Catalog / hardware
AI hardware
Local inference and homelab gear for agents that should not send every token to the cloud. Cloud models still pair well via cheaperllm and OneAIPass.
As an Amazon Associate we earn from qualifying purchases. Links use affiliate tag secondspring03-20. See full disclosure.
GPU workstations
Desktop GPUs for larger context and fine-tunes. Prefer cards with VRAM headroom for concurrent agent tools + model weights.
Local LLM hardware
RAM kits, NVMe, and cooling for Ollama / llama.cpp hosts. Pair with AgentStack MCP filesystem + sqlite for offline loops.
USB AI accelerators
Edge NPUs and USB accelerators for light inference on laptops without a discrete GPU.
Homelab NAS
Store model weights, session tapes, and crawl caches. Agents need durable disks more than flashy RGB.
Hybrid stack tip
Run drafts locally; escalate hard reasoning to Claude routes on cheaperllm. Edit in Cursor; chat across models on OneAIPass.