Building Infrastructure for AI and AGI.

A real-time synchronisation engine for sharing intelligence assets — graphs, tensors, documents and streams — across users and nodes. Operates across trust boundaries with cryptographic guarantees of integrity and lineage. Infrastructure for intelligent collaboration at scale.

A distributed data layer for intelligent content delivery. Synchronise complex data structures like streams, graphs and documents across a global network. Traditional CDNs replicate blobs, but when your data is more complex, the CDN model breaks down.

This is the universal problem of data-sharing, optimised. An intelligent CDN that syncs only what changes, to only where it's needed, with full control over who sees what. Delta-only propagation means hyper fast sync from node to node to the browser. Selective node networks mean data stays where you define, not scattered globally by default.

This is infrastructure for a distributed data layer without building it yourself.

Intelligent systems typically choose CPU or GPU — or, when combining both, face separate memory architectures limited by bandwidth. Apple Silicon changes this. Unified memory enables zero-copy between operation classes, plus iterative high-frequency feedback loops between CPU and GPU not previously possible.

No Apple Silicon-optimised databases exist. The architecture is new. We're building one.

A real-time stream of intelligence assets — insights, events, updates, inferences — flowing through your systems and out to others. Subscribe to feeds. Publish your own. Route intelligence where it's needed, secure what's sensitive.

This is actionable intelligence at scale. Not static knowledge bases — living feeds that update as the world changes. Less noise, more clarity.

Large models are expensive, opaque, and inflexible. Training costs scale with parameters. Relationships are buried in hidden layers. Updating means retraining. Graphs offer a different path. Smaller, specialised networks that capture dimensions within information through explicit relationships. Transparent, modular, adaptable. Update what changes without rebuilding everything.

If cheaper intelligence wins the long game, Graph Networks lead — order of magnitude savings while matching or outperforming larger models.

What happens when multiple specialised graphs work together? A graph for domain knowledge, another for context, another for memory, behaviour, personal profiles. Each captures a different dimension. Each does what it does well. How do they communicate? What gets shared? What's hidden? When do they converge?

Our research focuses on the interface — training intelligent communication protocols between graph layers. Keys, priorities, selective sharing. The applications for this are immense.

Cross-border intelligence represents a new frontier. Systems that span economic boundaries — supply chains, markets, federations of competing interests. Behaviour that is probabilistic rather than declarative. Outcomes that emerge from interaction, not instruction.

The question shifts from "what should the system do" to "what game are we designing". Game theory applied not as abstraction, but as architecture. Designing the rules so coordination compounds value, so sharing outperforms hoarding.

Schema is typically static — defined upfront, maintained separately from data, disconnected from how data actually behaves.

What if schema was learned? A statistical topology mapping term frequencies, co-occurrence, relationship distributions. Short codes for common terms, approaching Shannon entropy. The same mapping that optimises delta communication becomes a signal for retrieval — a query enters the structure, navigates encoded relationships, informs search before heavier operations engage.

Schema versions that evolve with the data it describes.

Rendering billions of nodes in the browser using Rust, WebAssembly and WebGPU. 3D visualisation for spatial navigation through complex structures. Multi-layered information sets that expand and collapse. Real-time data integration. Analytics capability on large datasets while navigating complexity without losing context.

What if the interface was a projection of the graph itself? UI dynamically generated from structure.

UI traversal is extended with graph inference — predictions on the next step, making for UIs that are highly intuitive, and adaptive, able to recognise patterns and integrate changing models in realtime. The UI learns alongside the user.