● Live now
AgentBOX, available today.
Shipping now: AgentBOX on the T2 Jetson Orin Nano Super as a single SKU, the Base membership, and the free Community path for hardware you already own.
Hardware · T2 · Launch SKU
AgentBOX
A pre-configured AI assistant on hardware you own. Ships with the MailBOX skill (inbox triage and drafting) and the OpenClaw conversational front-end (companion app, Telegram, WhatsApp, web) live out of the box. Includes a year of Base membership.
Shipping
Membership
Base
The managed first-party experience: pack allowance, OTA updates, idle-time intelligence, dashboard, and support.
Available
Hardware · T0 · Free
Bring your own
Run the full platform on a machine you already own, on the free Community plan. The open-core sovereignty path.
Available
Coming next
More metal, more membership.
The hardware ladder extends both ways — a low-cost dedicated node below the launch box, and serious headroom above it — while the membership ladder opens up multi-pack tiers.
Hardware · T1
Raspberry Pi / SBC
The cheapest dedicated, always-on node. A Pi-class board running thUMBox at a few watts for a private chatbot and light triage.
In development
Hardware · T3
Jetson AGX Orin
64GB, up to 275 TOPS — multi-pack hosting and larger models on the same NVIDIA stack as your launch box.
In development
Hardware · T3 · hardened
Mac mini M4
The x86 path with NemoClaw's kernel-level OpenShell sandbox — for when hardened isolation is a requirement.
In development
Membership
Plus
Multiple entitled packs (sequential hot-swap on T2, concurrent on T3+), deeper analytics, and more idle-time work.
In development
Membership
Pro
All packs, orchestration, fine-tuning, managed backup, and priority support — the full platform experience.
In development
On the horizon
Team and org scale.
When one operator's box becomes a team's shared brain — and then an organization's.
Hardware · T4
Multi-GPU workstation
64GB+ running 14–30B models with orchestrated multi-agent workloads — a single box a small team shares.
Planned
Hardware · T5
Rack / fleet
128GB+ running 30–70B models under fleet-coordinated policy — org-wide intelligence on metal you own.
Planned
Membership
Enterprise
Fleet management, org-wide policy, SSO, and a dedicated success path. Procurement and security on your terms.
Planned
The platform
A community-owned cloud.
The end state isn't more boxes — it's a network of them. Every thUMBox already has idle capacity. Pooled, that idle capacity becomes a cloud that nobody rents to you, because you and your peers already own it.
◇ Community cloud inference
A cloud, owned by the people in it.
Instead of renting inference from a hyperscaler, thUMBox owners pool spare capacity into a shared network. When your box is idle, it can contribute; when you need a burst beyond what your own hardware holds, you draw from the community — never from a data center.
- →Inference stays on community-owned hardware, end to end.
- →Capacity scales with the network, not a vendor's price sheet.
- →Your data-sovereignty guarantees travel with the work.
◇ Inference marketplace
Rent capacity. Or earn from yours.
A two-sided market for local-first compute. Draw inference from the community when you need it, or offer your box's idle cycles and earn from them. You set your terms, the network clears the price, and the economics flow to the people who own the metal — not a middleman.
- →Turn idle time into income — your box works while you sleep.
- →Burst beyond your tier without a cloud contract.
- →Transparent, peer-to-peer pricing — own the economics, literally.
Roadmap items beyond "Live now" are directional, not commitments — sequencing and specifics may change as we learn from the fleet. Hardware tiers and their capabilities are detailed on the Hardware page.