The Box — in development.
Today the cloud. Tomorrow the shelf. The Aetherframe you can use right now is hosted — powerful, private by policy, ready today. What's coming is the endgame: a small black box that runs the same thinking environment entirely on your own hardware, in your own home, with no path back to anyone.
Ships today
The hosted app, on Claude
Coming
The local-only box
Same vault
Same doctrine, zero cloud
Early access
Pro members first in line
Private by physics, not by policy.
The hosted Aetherframe is private by policy — your material lives in your account, gets cited back to its source, and gets purged when you say so. That's enough for most matters.
The box is for when it isn't. A small, quiet device that runs the same thinking environment entirely on your own hardware, on your own network, with no path back to us. Same vault. Same chronology. Same refusal to overstep. Zero cloud.
In development · Filter + capable local chat
Aether Core — the entry tier.
A small, silent, fanless ARM device — about the footprint of a hardcover book. Designed to run Filter Mode at line speed and handle conversational AI on 3B–7B local models. The privacy filter is the star at this tier; the AI is capable but not powerful.
For: privacy-conscious individuals, families, small offices. The household that wants the trackers blocked, the children's network filtered, and a sovereign AI to chat with — without sending anything to a cloud.
Target specs:
· ARM SBC, 8GB RAM, 256GB NVMe
· ~15W power draw, fanless, silent
· Dual interface networking or Ethernet + WiFi AP
· Llama 3.2 3B / Phi-3 Mini / Qwen 2.5 3B and similar
In development · Mid tier for serious work
Aether Pro — real reasoning, locally.
A mini-PC class device planned for strong CPU inference, optional integrated GPU, and enough memory to run 7B–14B models comfortably. The AI is designed to be good enough for serious professional work: legal document analysis, medical record review, financial modeling, evidence management.
For: professionals with sensitive data — attorneys, clinicians, therapists, financial planners, small-firm operators.
Target specs:
· Intel/AMD mini-PC, 32–64GB RAM
· 1TB NVMe, dual 2.5GbE networking
· Optional discrete or integrated GPU
· Llama 3.1 8B / Qwen 2.5 14B / Mistral 7B and similar
In development · High tier for the heaviest work
Aether Sovereign — frontier-class without the cloud.
A small-form-factor system planned around a dedicated GPU — enough VRAM to run 70B-class quantized models at speeds comparable to cloud APIs. This tier is designed to compete with paid cloud subscriptions, but with every prompt and every byte of context staying on the device.
For: sovereign tech enthusiasts, small firms, anyone who wants frontier-quality AI without the cloud terms-of-service.
Target specs:
· SFF system, 64GB RAM, dedicated GPU (24GB VRAM target)
· 2TB NVMe, dual 2.5GbE + WiFi 6E
· Llama 3.1 70B / Qwen 2.5 72B at production speed
· Local API endpoint for your own apps
Honest about the timeline.
We're not running fake countdown timers or selling pre-orders we can't fulfill. The box is in development. The hosted app, which uses the same software stack, ships today. Get on the early-access list and we'll tell you exactly when units are available — and what we learned getting there.