Customize your PAI Starter 32 GB Hearth
Accessible sovereign privacy. 32GB DDR5, whisper-quiet thermal design, and PAI OS.
Or $75/mo. for 12 mo. with 0% interest financing
Your Custom ConfigurationPAI-SPEC-2026
- Silicon & Memory32 GB Dual-Channel DDR5
- NVMe Storage1 TB PCIe 4.0 NVMe SSD
- Hearth Deck InterfaceLaser-Etched Monolithic Lid
- Power & Interconnect65W Monolithic GaN Adapter
- Software & WarrantyPAI OS Lifetime License
What's in the Box
Choose your finish.
Unified Memory.
Unified memory is shared directly between neural tensor cores and the CPU. High memory allows running large-parameter foundation models entirely in RAM at maximum token velocity.
NVMe High-Speed Storage.
Dedicated local storage holds your private Vector DB, model weight checkpoints, and indexed chat memories.
Top Deck & Interaction.
Choose the physical top interface for voice feedback, live agent telemetry, and tactile controls.
Power & Connectivity.
PAI Care & Software Support.
Guaranteed replacement coverage, direct engineer chat support, and curated monthly neural weights.
Founder batch units are limited to 250 serialized systems. Your reservation locks in the 60% launch discount and priority production slot.
Lock in 60% DiscountDetailed Architecture & System Overview
PAI Starter 16 GB is a small server, about the size of a Mac mini, that runs open AI models in your home. It is for people who want a private assistant without building one: it arrives with the PAI software on it, and the box is the whole product.
Inside is a Ryzen 8745HS/8845HS-class processor with Radeon integrated graphics, 16 GB of DDR5 memory, a 1 TB NVMe drive, USB4, 2.5G Ethernet and Wi-Fi 6. With 16 GB we plan for it to run 7-14B models and gpt-oss-20b. It cannot run every model: nothing home-sized runs the largest frontier models.
On first boot PAI interviews you for about 15 minutes. Then it writes a plain-English page of what it will and will not do. You can read it and edit it.
You talk to PAI in the PAI app, an installable web app that only your own devices reach over your private network. Signal and Matrix, both end-to-end encrypted, are planned for month 1. Telegram and email work at launch, with a catch: Telegram relays bot chats on its servers and is not end-to-end encrypted, and your mail provider carries your email. WhatsApp and iMessage are optional and unsupported. Inference is local. If you choose WhatsApp or iMessage, Meta or Apple still carry the message.
No PAI account holds your data, and telemetry is off by default. Your personal memory and the search index over your own files live in a SQLite file on the box. Remote access goes through a private overlay network, so nothing listens on public interfaces.
The Starter is built to grow. Step one is more memory: the box has two memory slots and the second is left open. One memory stick runs in single-channel mode, which is slower until the second slot is filled. Step two is a GPU dock. Step three is a second box for bigger models. The dock port and the slot count are still to be confirmed on the supplier unit.
We have not measured speed, power or noise yet. We will publish measured numbers for each model and each upgrade stage after prototype testing. Which models fit on which memory size is an estimate until then.
Looks small. Lives large.
Five by five inches of sovereign AI compute. Precision CNC titanium unibody, NVIDIA-grade cellular acoustic mesh, and unified high-bandwidth silicon.
Bead-blasted champagne titanium unibody with 45° chamfers and elevated perimeter base.
Sovereignty from the inside out.
Engineered for continuous 24/7 autonomous agent execution. Zero noise at idle, ice-cold under continuous inference.
Titanium Monolith
Machined from a single block of aerospace-grade aluminum alloy with silky bead-blasted texture and diamond-cut 45° chamfers.
Dual Radial Blowers
Twin curved-blade radial fans pull fresh air across the 360° perimeter foot and channel it across the thermal core at near-zero dB.
Copper Vapor Chamber
Sintered copper vapor chamber contacts 100% of the unified silicon die, dissipating sustained 30W-50W workloads without throttling.
Shielded Sub-Frame
Isolates vibration and grounds heavy I/O lanes including rear OCuLink 64Gbps, dual 2.5GbE, dual Thunderbolt 4, and HDMI 2.1.
The Hearth Display Concept.
Exploring integrated top-mount OLED displays for the top two PAI models. Glanceable agentic workflows, real-time voice waveforms, and local touch control directly on the hardware.
A panoramic top display integrated into the obsidian-black chassis. Visualizes live token velocities, neural memory graphs, and parallel sub-agent workflows.
Edge-to-Edge Rectangular AMOLED HUD Deck
Live Agentic Orchestration
Follow multi-step autonomous agent tasks in real-time: research synthesis, database queries, and code builds.
Real-Time Neural Telemetry
Monitors unified 128GB memory bandwidth, token generation velocity (tok/s), and active KV-cache headroom.
Standalone Voice & Touch Hub
Operate your Sovereign Hearth directly without needing an external studio monitor or waking your laptop.
Specs
- Processor
- Ryzen 8745HS/8845HS-class, Radeon integrated graphics
- Memory
- 16 GB DDR5, two SO-DIMM slots, second slot left open for upgrade (slot count to be confirmed)
- Storage
- 1 TB NVMe
- Connections
- USB4, 2.5G Ethernet, Wi-Fi 6
- Runs (planned)
- 7-14B models and gpt-oss-20b
- Upgrade path
- More memory, then a GPU dock (port to be confirmed), then a second box
- Size
- About the size of a Mac mini
- Certifications and warranty
- To be confirmed
Measured numbers coming after prototype testing
Upgrade path
Start here
More memory
Fill the open second slot. One stick runs single-channel until both are filled.
Then
GPU dock
An add-on graphics dock. The port is to be confirmed on the supplier unit.
Then
Second box
Add another PAI box for bigger models.
