Sovereign AI
The enterprise owns the intelligence it builds: its data, its models and the rules that govern them.
Sovereign deployment stack
Every layer of a Kamino deployment runs inside the client's environment.
Principles of a sovereign deployment
Ownership of the enterprise context
Source data, the Tabula model and the logic built on them are client assets, governed as objects, actions and permissions the client controls.
- Durable enterprise knowledge held apart from transient prompts
- Every workflow mapped to its accountable owner
- The model exportable in standard formats
Model independence
Open-weight models are served on compute inside the client environment. Each task is routed to an appropriate model, and models are added or replaced without moving data.
- No dependence on an external AI provider
- Prompts, records and outputs remain in the environment
- Model changes require no change to business logic
Governance of every action
Client rules decide what agents may read, propose and change. Actions with external effect wait for a named approver.
- Gateway-mediated access, read-only by default
- Every request, source and decision recorded
- Any run may be halted by an authorised user
Deployment within the client environment
Kamino installs the platform on premises or in the client's private cloud tenancy. Software arrives as signed releases, and no channel carries client data outward.
- On-premise, including model serving
- Private cloud, under the client's account and region
- Each deployment a dedicated installation
Any data and any model, on the client's infrastructure
Self-hosted open-weight models serve the client's data where it already resides.
Replaceable models, retained context
Providers change without moving data or rewriting the logic built on it.
One-way software delivery
Kamino delivers signed releases into the environment and holds no client data.
Deployment specification
Each component of a Kamino deployment, where it runs, and what it holds.
| Component | Runs in | Holds | Network |
|---|---|---|---|
| Model serving | Client data centre or private cloud, GPU nodes | Open-weight model weights | Internal only |
| Metron runtime | Client environment | Run records, audit trail | Internal only |
| Tabula | Client environment | Unified business model | Internal only |
| Secure gateway | Client network, in front of systems of record | Request log | Systems of record only |
| Software releases | Built and signed by Kamino | No client data | Inbound only |
Data residency
The location of every class of data in a Kamino deployment.
| Client environment | Kamino | |
|---|---|---|
| Source data and systems of record | None | |
| Tabula model | None | |
| Metron run records and audit trail | None | |
| Prompts, retrieved records and model outputs | None | |
| Model weights and inference | None | |
| Software releases | Installed | Built and signed |
Enterprise AI should answer to the organisation that owns the data.
Approach→