Three products, one set of foundations
Soika Stack is the inference foundation: it runs models on your GPUs and can serve them to anything above it, including Soika AI and Mockingjay. Those two are separate platforms with separate purposes and no dependency on each other. What they share is how they deploy — the same sovereign options, the same infrastructure and the same governance bar.
One foundation. Two independent platforms.
Buy one, buy all three. Soika Stack can sit underneath either platform as the model-serving layer, or underneath your own applications. Soika AI and Mockingjay never require each other.
The hardware underneath, specified properly
Software performance is decided long before the first token. Soika builds and ships its own AI workstations, laptops and cluster designs — every system arriving with the Enterprise licence, Soika Stack and Mockingjay already installed.
AI workstations
Desk-side development and fine-tuning systems for data science teams, pre-configured with the Soika toolchain.
GPU servers
Dense inference and training nodes sized to your workload, from single-socket L40S boxes to 8-way H200 systems.
HGX clusters
Reference architectures for multi-node HGX and GB200 deployments with validated fabric and power design.
Storage & networking
High-throughput parallel storage and InfiniBand/RoCE fabrics engineered for sustained inference and retrieval.
Deployment models for organisations that cannot compromise
Choose the boundary that matches your regulator, not the one that matches a vendor's business model.
Sovereign cloud
Deployed into a national or regional cloud with in-country data residency, local operators and local support.
- In-country residency
- Local operations
- Regulator-aligned
Private cloud & on-premise
Runs in your own data centre or VPC on hardware you own, integrated with your identity and network controls.
- Your hardware
- Your identity provider
- Your network policy
Air-gapped
Installed from signed offline media with no outbound connectivity, for classified and critical environments.
- No egress
- Offline updates
- Signed artefacts
The same controls in every product
Identity, policy, evidence and residency are not re-invented per product. Whichever combination you deploy, the governance model your auditors review is the same one.
Identity & access
SSO, SCIM and role-based access down to individual models, agents, tools and datasets.
Audit & evidence
Immutable logs of every model call, retrieval, tool invocation and agent decision — exportable for regulators.
Evaluation & drift
Golden datasets, regression gates and continuous quality monitoring before and after every release.
Policy enforcement
Declarative guardrails on data access, spend, autonomy and approvals — enforced ahead of the action.
Data protection
PII detection, redaction, retention policies and residency pinning applied consistently in every product.
Isolation
Hard multi-tenant boundaries so departments, agencies or customers never share context or capacity.
Pilot in weeks. Scale on evidence.
We do not start with a platform rollout. We start with one process that matters, then earn the right to expand.
Discovery
A structured workshop maps candidate workflows, data sensitivity, existing systems and the governance bar you must clear.
Architecture
We size the GPU estate, define the deployment topology and design the agent and fleet model against real processes.
Pilot
One department, one measurable outcome, in your environment — with evaluation baselines agreed up front.
Scale
Replicate proven fleets across regions and business units, with local policy overrides and central observability.
See it against your own architecture
Bring your current estate, your constraints and one process you would like to automate. We will show you what the deployment actually looks like.