Structure as guarantee.

I. A routing and protocol layer for AI systems.

Sonata42 began with a question about scale. Its 2024 proposal, NeuroNet Synergy, set out the hypothesis that smaller, connected AI systems, coordinated through established network protocols, could outperform large single models in efficiency and scalability without a meaningful loss of accuracy.

The open problem in that approach is capability discovery: how one system finds the specialist system, possibly itself generative, that can best answer a query. Retrieval augmented generation and the Model Context Protocol both point at this need without solving it. Internet routing offers a working precedent. Border Gateway Protocol, Open Shortest Path First, and Multiprotocol Label Switching let traffic flow between any two endpoints although no participant holds a global view. The planned elements of the routing layer are a capability advertisement format, a route selection method, a session handshake, and a label switched forwarding plane.

II. Measurement integrity: generated values must not pass as measured ones.

Generative AI is entering settings where the difference between a value that was measured and a value that was generated matters. Asked for the height of tide in a harbor, a model may produce a plausible figure from its training data instead of computing it from the official tidal data. A sensor reading thirty seconds old looks the same as a live one. Asking an AI system not to invent figures is not enough. The guarantee must be structural.

Some functions in a technical system, such as reading sensors or performing known calculations, are deterministic. Others, such as planning, fault diagnosis, and recording observations, can benefit from generative assistance. The measurement integrity layer holds the two apart by construction. Information flow labels, in the manner of the LLMbda calculus, attach source and age to every critical value. A gate reports a value only when it has been measured or retrieved, and otherwise the system says it cannot tell and sends the operator back to the primary sources. Agents are composed with the plumbing calculus, a typed language with category theoretic foundations, whose compiler checks that an agent network is well formed before it runs.

The target application areas are systems where a wrong value has real consequences, such as maritime, energy, and industrial systems. These are targets for the research. Sonata42 has no deployments in these areas at present.

III. How the two connect.

Both lines of work treat the coordination of AI systems as a protocol problem rather than a model problem. The routing work asks how a query reaches the right system. The measurement integrity work asks what must travel with each answer, its source and its age, so that a generated value cannot be mistaken for a measured one. The commercial aim is a provenance gated runtime for AI agents in critical operational technology.

IV. A current project: Slot Gateway.

Slot Gateway is an early stage project within the routing and protocol line. It aims to give concurrent AI agents fair, exclusive, time-slotted turns at shared tools, through an open source gateway released under the Apache 2.0 licence. Read more on the Slot Gateway project page.