Everything that runs Agent One.
We don't hide what One is made of. This is a published bill of materials for the 🤫 Agent One Platform and the 🤫🐶 One Supercomputing Platform — every model, agent runtime, silicon line, cloud, developer platform, identity and payment rail, and open standard we use to do the job for our customers, alongside the components hussh builds. Used under agreement with our partners.
The software that thinks, decides, and acts.
The intelligence stack, top to bottom — from the models to the rails of real life.
Models
Agent One is model-agnostic — bring the mind you trust. We integrate the frontier so your One runs on the best available.
Frontier language and reasoning models
Frontier reasoning and agentic models
Frontier multimodal models
Frontier models for real-time reasoning
Open-weight Llama models for on-device
How One acts, not just answers
Agent One runs on and interoperates with the latest open agent runtimes — the harnesses that turn a model into an agent that takes actions, on a schedule, with memory — across both consumer and enterprise editions.
Nous Research agent runtime — Agent One's reasoning loop
Agent Development Kit — agentic orchestration in our backend
Open self-hosted agent runtime — interoperates via open standards (MCP)
Frontier agent tooling, reached through MCP and A2A
On-device & serving
The layer that runs intelligence efficiently on owned silicon, then routes overflow to where it runs best.
On-device inference on Apple silicon
GPU acceleration on NVIDIA silicon
High-throughput model serving
Portable, cross-hardware model format
Orchestration for hybrid and burst workloads
Compute on your bill
When a job outgrows the device, One bursts into the cloud you own — provisioned, run, and torn down, keys never persisted.
Sovereign cloud and enterprise burst
Elastic accelerators and storage
Accelerators and managed services
OCI — large-scale AI compute capacity
GPU-specialized cloud for burst
Where work already lives
One meets your software where it is, integrating through the platforms enterprises already run on.
App distribution and platform integration
Android and Workspace integration
Windows and Microsoft 365 integration
CRM and customer-data integration
API integration across enterprise systems
The rails of real life
Sign in with the identity you already use — including the ones US government agencies already trust — sign what needs signing, and move money on rails you already trust, every step consent-gated.
Out-of-the-box: government-grade verified identity — US tax citizens, employees & veterans supported by default
Out-of-the-box: the US government's own sign-in for federal services
Standard, out-of-the-box account connection — by consent, with a receipt
Biometric identity on Apple devices
Account identity on Android and the web
Agreements and e-signature
Secure account and banking connectivity
Person-to-person money movement
Agent & commerce rails
Open, published protocols so agents and institutions plug into One without a walled garden.
Model Context Protocol — tool and data access
Agent-to-agent coordination
Agent payments protocol
Universal commerce protocol
hussh's consent & transparency protocol
The hardware that runs it — owned by you.
One Puppy is the owned-silicon line, from the phone in your pocket to a deskside supercomputer.
Hardware you own
One Puppy spans owned silicon from pocket to deskside — unified-memory and accelerator hardware tuned for local intelligence.
Unified-memory silicon, iPhone and Mac
Accelerators and DGX-class deskside supercomputers
On-device NPUs for mobile
CPU/GPU compute for workstations
TPU-class accelerators for burst
Android devices and memory
The CPU architecture under Apple, Qualcomm & Grace
x86 compute and an emerging foundry
The real bottleneck of the AI era
High-bandwidth memory (HBM) is the single scarcest input in modern accelerators. We track it as a first-class supply-chain risk.
HBM leader — the binding constraint on accelerators
HBM, DRAM, and NAND flash
HBM, DRAM, and storage at scale
High-capacity storage
NAND flash supply
Moving data between chips and boxes
Fleets of Puppies are only as fast as the fabric between them — interconnect is its own supply chain.
NVLink and InfiniBand (Mellanox) interconnect
Ethernet switch silicon and custom accelerators
Data-center class switching
Custom silicon and optical interconnect
Who builds the boxes
One Puppy is assembled by the system builders and ODMs who turn silicon into shippable supercomputers.
AI server systems and deskside builds
Workstation and server integration
Devices and edge systems
Contract manufacturing at scale
ODM server manufacturing
Upstream dependencies we track for resilience
These are not direct commercial partners — they are the upstream chokepoints every accelerator depends on. We name them because understanding concentration risk is how we stay resilient.
Leading-edge foundry for nearly all AI silicon
Alternate leading-edge fabrication
Emerging US-based leading-edge capacity
EUV lithography — the chokepoint above the foundry
Fab process equipment
Etch and deposition equipment
Where Puppy is going — a distributed supercomputing grid.
The north star: 🤫 Puppy One running in garages and warehouses, on Starlink, fleet-managed by Agent One — sited where power is cheap or subsidized, cooling is abundant, and sunshine is maximal. Several pieces below are on the roadmap, not shipped.
Backhaul for compute that lives anywhere
The distributed grid puts Puppies in garages and warehouses and connects them to the world — satellite-first, terrestrial where it's better.
Satellite backhaul for off-grid garages and warehouses
Terrestrial mobile and fiber backhaul
5G connectivity and edge reach
Mobile and fixed-wireless backhaul
Edge network, peering, and security
Lowest cost per watt wins
We site compute where electricity is cheap or subsidized and sunshine is abundant — and pair it with on-site solar and storage.
Solar and Powerwall / Megapack storage
Microinverters and home/garage solar
Siting in cheap-power, high-sunshine regions
Heat is the enemy; cooling is the moat
Distributed compute lives where cooling is abundant — naturally cool sites plus liquid and immersion cooling for density.
Thermal management and power distribution
High-density cooling for garage and warehouse nodes
The network of operators who host and maintain the grid
The parts that are ours.
The components we design and own — the agent, its circle, its engine, its consent ledger, and its hardware line.
🤫 Agent One
The personal agent that remembers, decides, and acts — loyal to one party: you
The agent circle
Kai, Nav, and the specialist agents, each with one job
One Engine
Benchmarks your machine and routes each workload on-device, hybrid, or cloud
PCHP consent ledger
Every access consent-gated and written to a transparency log only you can read
🤫🐶 One Puppy
The owned-hardware line that brings the supercomputer into your hands
Honest about what this is.
One is model-agnostic — you bring the AI you trust, and the components above are how we make it run on hardware you own. Specific models, silicon, memory, and configurations are confirmed at delivery, and some capabilities are on the roadmap.
These technologies are used under agreement with our partners. All company and product names and logos are trademarks of their respective owners.
Partners vs. dependencies. The lists above include both direct partners we build with and upstream supply-chain dependencies — foundry, lithography, memory, connectivity, power, and cooling — that we track for concentration and resilience risk. Naming a company does not imply a commercial relationship; it reflects, in the open, what the platform actually depends on.
On compliance, we keep our word: FedRAMP High and supporting authorizations are in active pursuit. We say authorized only when we are.
Built with the best. Owned by one.
The finest AI-first, human-first, supercomputing-first technologies on earth, assembled into an agent that answers only to you.
One is a product of Hushh Technologies Corporation (brand: 🤫 “hussh”), an independent company. One runs on third-party silicon, systems, and cloud; all company names are used solely to describe the platforms on which One software runs. Hushh Technologies is not affiliated with, endorsed by, sponsored by, or partnered with any company named.