Contextual interoperability

Context is only yours if any model can use it

Contextual interoperability is what lets you use your own notes, frameworks, and commitments with any AI model a task calls for. It has two halves: the context has to be legible, with its relationships explicit enough for a model to follow, and it has to be portable, held by you and lent to a model for a task instead of living inside one provider’s system. Each half is of little use without the other.

Contextual interoperability

One-sentence definition: Contextual interoperability is the property of personal or institutional context that lets any AI model, local or hosted, read it and reason within it, without the context losing its meaning or leaving your control.

Interoperability usually describes systems that can work together, such as a hospital’s records system exchanging data with a GP’s. Here the thing that has to work across systems is your thinking. Most people’s context falls short in one of two ways: it’s either a pile of documents a model can only skim, or it’s built up inside one product’s memory and can’t be taken anywhere else.

Legible: relationships a model can follow

A model given a folder of notes sees disconnected text. It can find passages that match a question, but it can’t see that one framework extends another, or that a study you read last year undermines an approach you still use. Those relationships usually sit in your head. Making them explicit, as typed links in a knowledge graph or clear structure in plain text files, turns a store of notes into something a model can reason across. The post Organising your notes for AI works through how.

Portable: context that stays yours

Context can be well structured and still sit inside one vendor’s system, where only that vendor’s model can use it. Portability means the context lives where you control it and a model is applied to it for a particular task: a local model¹ for sensitive material, a more capable hosted one for demanding work, and a different one next year without rebuilding anything (Rowe & Lynch, 2025). Model Context Protocol is one open standard for connecting models to context in this way, and intelligence as a service describes the arrangement that makes it safe, in which you use the model without the provider keeping your context.

In a programme team

Take a pharmacy programme whose curriculum map records, in plain text, which learning outcomes each module assesses, which simulation sessions prepare students for which OSCE² stations, and where the programme has decided to accept a gap. A model reading that map can answer “what breaks if we move this module to year three?” by following the relationships, where a model given the module handbooks as PDFs can only summarise them. Because the map belongs to the programme team, the same file can go to the university’s approved model when student data is involved and to another tool for drafting, and it outlasts any one contract.

This is what makes context sovereignty practical, since owning your context doesn’t help much if no model can read it, or if only one provider’s can.

Sources

• Rowe, M., & Lynch, W. (2025). Context sovereignty for AI-supported learning: A human-centred approach (Version 0.9) [Preprint]. OSF. https://doi.org/10.31219/osf.io/8czva_v2