General
What is AgentStation? +
A purpose-built computer for running AI agents locally. Pre-configured with everything you need to run large language models without cloud services, subscriptions, or data leaving your hardware.
How is this different from building my own PC? +
Three things: (1) Optimized component selection for AI inference. (2) Pre-configured OS, GPU drivers, CUDA stack, agent templates. (3) Ongoing support and software updates. DIY takes 10-20 hours of research and debugging.
Do I need internet? +
Only for downloading new models and updates. Once set up, runs 100% offline. Ideal for air-gapped environments (HIPAA, defense, finance).
Performance
What models can I run? +
Core: 7B-13B natively. Pro: 7B-13B native, 70B with offloading. Max: up to 180B with quantization. Llama 3.1, Mistral, Qwen, CodeLlama, Stable Diffusion XL, and more.
How fast is inference? +
Core: ~25 tok/sec (8B). Pro: ~45 tok/sec (8B), ~15 tok/sec (70B). Max: ~80 tok/sec (8B), ~35 tok/sec (70B). Human reading speed is ~4 tok/sec.
Can I upgrade later? +
Yes. Standard ATX components. Swap GPUs, add RAM, expand storage. Max has 80 PCIe lanes for dual GPUs.
Ordering
When will my order ship? +
Pre-orders ship Q2 2026 (April-June). Express builds: +$199 for 2-week turnaround.
What is the refund policy? +
Full refund anytime before shipping. 30-day return after. 2-year hardware warranty included.
Do you offer volume discounts? +
5+ units: 10% off. 10+: 15% + dedicated support. 50+: custom pricing. Email sales@agentstation.co.