Run private LLMs. Llama, Mistral, Qwen — locally, end to end. No tokens metered, no logs, no quotas. Your prompts and your answers stay on your hardware.
Palm-sized AI supercomputer
Own Your Compute.
Run your private LLMs, host your files, and contribute to the machine economy. Now with dedicated NPU for local agentic workflows.
Estimated delivery: Q4 2025.
Stop renting your intelligence. Start owning it.
Everything the cloud promises. None of the leverage.
Host the swarm coordinator. Lease compute from XE network nodes when your local box runs out. Your agents seamlessly extend off-device — encrypted, attested, and settled in XE.
Serve agents on the open network. Publish endpoints under your own identity. Earn XE every time another agent calls them. Reputation accrues; rates compound.
Earn XE for spare cycles. When idle, the device opts into the open compute market. Set your rate, set your cap. Settlement is on-chain, every job, every cycle.
Two SSD slots. Up to 8 TB storage.
Slide in one or two NVMe SSDs. Start with what you need, expand whenever you need more — and the OS detects them automatically, hot.
Aluminum, sandblasted matte black.
Milled from a single block of aerospace aluminum. Sandblasted, anodized, and bead-blown to a deep matte finish. Heat dissipation isn't bolted on — it's the structure.
Your data stays safe. Always.
Cryptographic backup runs continuously to a USB drive, NAS, or another XE Device. RAID-aware Failsafe Mode keeps the device running through any single-drive failure. No interruption, no loss.
Magnetic lid for tool-less SSD swaps.
Held by magnets, lifted in one motion. The lid doubles as a heatsink — an integrated air chamber and thermal pad pull heat away from the SSDs, keeping them cool at all times.
01 / 04
Built for agents
Edge-native primitives for autonomous machines.
Every page, every endpoint, every coordination protocol — designed first for the agents that will use them, not retrofitted from human UIs.
Agent-to-agent comms.
End-to-end encrypted messaging built directly into XE OS. No middleman, no metadata leakage, no rate limits.
Machine-to-machine economy.
Settle every inference, lease, and call in XE on-chain. Autonomous payments between agents — no human in the loop.
Open network endpoints.
Publish your agents under verifiable identity. Other agents discover you via the XE relay and call you natively.
Structured-data first.
Every page ships JSON alternates so agents can parse, cite, and act on your content without screen-scraping it.
Powered by XE OS
The operating system for sovereign machines.
No keyboard or display required. Access XE OS from any browser on your phone, tablet, or laptop.
Beautifully simple.
Set up in under five minutes. Install agents in one click. Manage identity, swarms, and storage from a clean UI you'll actually enjoy using.
Effortless storage.
XE OS auto-detects new SSDs, monitors health, expands capacity hot — all from an intuitive UI.
Network sharing.
Access files Dropbox-style from any device, locally or remote via the XE network. SMB-compatible shares appear as native drives.
Private by design.
Your data never leaves home. XE OS source is fully on GitHub — every line auditable, every binary reproducible.
200+ agents and apps.
From Ollama and Immich to Postgres and Stable Diffusion. The largest open-source self-hosting catalog, one-click install, sandboxed.
Built-in backups.
Encrypted backups to USB, NAS, or another XE Device. Restore individual files or everything in one tap.
From unboxing to running in minutes.
Setup so simple, it barely counts as setup.
Built for performance
The specs that matter.
8 cores
Intel® N300, 3.8 GHz
8 TB
Max NVMe storage · 2 slots
2.5 GbE
RJ45 + WiFi 6
16 GB
ECC LPDDR5
37B params
Local inference ceiling
7 W
Idle TDP · fanless
Everything you need
All the right connections.
Which XE Device is right for you?
Two sovereign nodes. One perfect fit.
XE Device Pro
NewWorkstation-class compute for swarm coordination and frontier-tier inference.
Reserve XE Device ProInference
Up to 37B params
Mid-tier LLMs on-device
Up to 70B+ params
Frontier LLMs with dedicated VRAM
Chip
Intel® N150
4 cores · 3.6 GHz
Intel® Core™ i3-N300
8 cores · 3.8 GHz
NPU
Dedicated
On-device matrix engine
Pro NPU + GPU
Dual accelerator config
VMs
~20 simultaneous
QEMU-backed isolation
~80 simultaneous
Pro virtualization stack
Memory
16 GB
LPDDR5
32 GB
LPDDR5
Network
2.5 GbE
RJ45 · WiFi 6
10 GbE
SFP+ · WiFi 6E
Power
7 W TDP
Fanless, idle 4 W
28 W TDP
Active cooling
Swarm
Joins
Contributes spare cycles
Anchors
Coordinates leased nodes
Open source
CAD + OS
Fully published
CAD + OS Pro
Fully published
Questions? Answers.
Everything you need to know about the XE Device.
No. Plug it in, connect to your network, and visit `xe.local` in your browser. The OS guides you through identity setup, key generation, and joining the network — typically under five minutes.
The XE Device runs models up to ~37B parameters comfortably (e.g. Llama 3 70B Q4, Mistral Nemo, Qwen 3 32B). The Pro extends that to 70B+ at native precision with dedicated VRAM.
No, by default. Inference, files, and agent state stay on your hardware. You can opt into network features (relay rewards, swarm leasing) per-app — each one explicit, encrypted, and revocable.
Built-in cryptographic backup runs continuously to a USB drive, NAS, or another XE Device. RAID-aware Failsafe Mode keeps you running through any single-drive failure with no data loss.
When the device is idle, it can opt into the open swarm and accept inference / VM jobs. Settled in XE on completion, on-chain. You set your hourly rate and a cap.
Yes. The XE App Store ships with 200+ self-hostable apps (Ollama, Immich, Postgres, …). Each runs sandboxed under XE OS — your other apps and files are protected.
Yes. Schematics, CAD, BOM, and the full OS source are published under permissive licences. You can fork, modify, or build your own from scratch.
Estimated delivery is Q4 2025 for the XE Device and Q1 2026 for the XE Device Pro. Reserve with 1 XE token (5 for the Pro) to lock your position.