TrustTwin OS
Governed compute for large, valuable datasets. Private AI, 3D and heavy processing on your nodes, under your policy.
Built in Europe with policy-grade control over data residency, inference and execution.
Making heterogeneous, partly decentralized compute reliably predictable for AI, 3D and remote-visualization workloads, under hard privacy, residency and quality constraints, is an unsolved problem. Cloud schedulers, FinOps tools and distributed compute marketplaces each solve part of it, and never all of it together.
One repeatable pipeline, from local acquisition to tools.
Data stays where it should. Work moves to the machine that is allowed to run it.
-
01
Local acquisition
Where the data is born: scanners, company machines, archives, local models.
-
02
Upload
Only what must move, and only what policy allows.
-
03
Distributed processing
Unattended, on nodes chosen by the router.
-
04
Upload
Validated results, with a record of where they ran.
-
05
Tools
Hubs, viewers, apps and agents, on any device.
Governed end to endPermissionsResidencyCapacityCost
The same five steps run a museum and a small company today.
How the Smart Compute Router works
At every step of the pipeline, each workload passes through six governed stages before, during and after execution.
-
01
Classify
Workload type and data sensitivity.
-
02
Constrain
Residency and security policy applied as hard limits.
-
03
Preflight
GPU, VRAM, CUDA, network and minimal execution verified before committing.
-
04
Route
Cost, time and risk optimised across local, cloud, edge and distributed capacity.
-
05
Recover
Retries, checkpoints and fallback when a provider fails.
-
06
Validate
Output completeness and quality checked, every decision logged.
The innovation is not cheaper GPUs. It is trusted compute governance.
What the router makes possible.
Four capabilities, the same for a museum and for a small company.
Heavy processing at a fraction of the cost
Unattended, on our own and partner nodes, at a fraction of hyperscaler cost. Customers can make their computing capacity available to the network when they do not need it, and unit cost falls as the network grows.
Complex workflows on local nodes
Any processing workflow, even a complex one, is set up quickly on the node network through the job scheduler. For every step, the router picks the machine allowed to run it.
Local AI, easily managed
Models run on your local nodes and we keep them current. Nothing leaves without policy.
Tools and real-time access on any device
Heavy data streamed from remote nodes to hubs, viewers and apps on any device, without the source data ever being copied to user devices.
What each layer does
Built as one system, not assembled after the fact: from context and policy down to validated, auditable output.
SweetHive Context Graph
Hive, context and group model. Every item is scoped at creation; apps and agents inherit the same boundaries.
Agent Runtime & Token Layer
Private agents, MCP-compatible access, revocable scoped tokens, audited actions. Includes Agents Node for on-device inference and distributed worker nodes.
TrustTwin Asset Model
Metadata, 3D files, condition reports, rights, lifecycle, provenance and remote visualization of heavy twins.
Data Sensitivity & Residency Engine
Classifies sensitivity and enforces local, private, EU-only or country-specific processing.
Preflight & Compatibility Engine
Verifies real capacity before committing a workload.
Smart Compute Router
Normalises provider offers, estimates cost and time, selects resources under hard constraints.
Resilient Execution & Fallback
Retries, checkpoints and alternative routing.
Output Validation & Evidence Layer
Validates results and stores decision history for audit.
TrustTwin OS is not a roadmap.
It runs today, in two products serving real customers.
AerariumChain came first, SweetHive followed. TrustTwin OS Core, for partners and enterprises running their own nodes, comes next.