DEPLOYABLE INFRASTRUCTURE
CPU-first. Sovereign-deployable.
Production-ready on commodity hardware.
The SurfaceOS platform is designed for deployment where GPU clouds are impractical, unaffordable, or unacceptable. It runs on standard server hardware, deploys in Docker or Kubernetes, and supports on-premises, edge, and air-gapped sovereign environments.
The Geode
The deployable unit
A Geode is the complete SurfaceOS deployment bundle: the SurfaceOS runtime, Exaline controller, and a bound AI model—packaged as a single deployable unit in Docker or Kubernetes. The CREnode specification defines the minimal production Geode.
| Model | Granite 4-nano (Q4 GGUF) |
| Compute | 1 vCPU |
| Memory | 1 GB RAM |
| GPU | None required |
| Deployment | Docker / Kubernetes |
| Inference | CPU-only |
| Network | Air-gap compatible |
This is not a development profile or proof-of-concept. It is a production execution unit deployable at scale on commodity infrastructure, cloud VMs, or edge hardware.
Benchmark
SurfaceOS vs. frontier LLMs on multi-step accuracy
SurfaceOS accuracy at 500+ sequential instructions
Frontier LLMs (GPT-4-class) at same instruction depth
Frontier LLMs degrade because extended instruction sequences approach or exceed effective attention window capacity, and probabilistic inference compounds error with each step. SurfaceOS does not rely on the model's context window to maintain state. State is managed externally in Deep Store. Each Geode receives only the bounded, scoped input it needs.
The 17x power efficiency advantage over GPU-based inference is a consequence of the same architectural separation. Small, bounded Geodes executing scoped tasks on commodity CPUs cost a fraction of frontier model API calls on every step.
Scale
Multi-Geode cluster execution
SurfaceOS supports multi-Geode cluster execution for complex workflows requiring parallel reasoning paths. The PlanGraph algebra defines formal cluster transition semantics: how parallel sub-graphs are dispatched, how outputs are validated, and how results are merged back into canonical state.
This enables SurfaceOS to scale complex workflows horizontally across multiple Geodes—each executing a bounded sub-task—while Exaline maintains global execution integrity. The cluster behaves deterministically from the outside regardless of which Geodes are active.
Deployment
Deployment environments
Enterprise On-Premises
Full SurfaceOS stack deployed within the customer's data centre. No external API calls. No data egress. Full audit log retained internally.
Sovereign Cloud
Deployment in a government or regulated-sector cloud environment. Compatible with NCSC cloud security principles. No dependency on US hyperscaler infrastructure.
Air-Gapped
Complete offline operation with no network connectivity required after initial deployment. Designed for defence, intelligence, and critical infrastructure.
Edge
Geode deployment on resource-constrained edge hardware. CREnode specification: 1 vCPU, 1GB RAM, no GPU.
Integration
How to integrate
SurfaceOS is designed as a layer beneath existing products, not a replacement. Integration paths include:
- OEM embedding of SurfaceOS runtime within an existing enterprise AI product
- Exaline as a supervisory orchestration layer above an existing model stack
- Geodes as bounded execution units within an existing agentic workflow
- Governance and replay layer as a standalone compliance module
- Protocol licensing for platform players requiring deterministic execution semantics