Connector

Two programs rarely speak the same language on their own. A calendar stores appointments differently than a spreadsheet, an online shop manages orders differently than an accounting system. A connector is a ready-made link that handles this translation. It fetches data from one program, brings it into the right form, and delivers it to the other. The comparison to a travel adapter for power outlets fits well: the outlet in Italy stays Italian, but your charger still fits into it. Importantly, a connector doesn’t invent anything new. It only links things that already exist.

Why AI is of little use without a connection to the outside world

A language model like ChatGPT only knows, on its own, what was in its training data. It knows nothing about your exam, your bank balance, or a company’s orders from last week. But this exact knowledge is what makes AI valuable in everyday work. For the model to access it, it needs a way out. Connectors are that way.

For companies, this is a matter of cost. Programming every connection yourself takes weeks and has to be maintained continuously. If a provider changes its interface, the self-built solution breaks. Ready-made connectors are handled by the vendor, maintenance included. That’s why press releases from software companies often advertise how many connectors a product comes with: it’s a measure of how quickly customers can get started.

There’s also a downside. Anyone who activates a connector gives a program access to sensitive data. A poorly secured connector can therefore become a gateway for attacks. That’s why experts pay close attention to exactly which permissions a connection really needs.

What actually happens during data exchange

A connector first logs into the target system, usually with a digital key instead of a password. It then requests data via a defined interface. Such interfaces are called APIs: a kind of order form that one program offers to other programs. The connector makes the request, receives the response, and converts it into a uniform format.

With AI applications, there’s an additional step. Long documents are broken down into small sections and stored in a way that lets you later find the passages that match in content. When someone asks something, the system searches out the matching sections and presents them to the model to generate an answer. This process is called Retrieval Augmented Generation, or RAG for short.

Connectors work in two directions here. Reading, they fetch information; writing, they create something, such as a new entry in the customer system. So that not everything has to be transferred anew with every request, many connectors keep track of what has changed since the last time. A common misconception is that a connector is a copy of the data. Usually, everything stays in the source system, and the connector only fetches it when needed.

Connectors in products and headlines

You encounter this principle constantly in everyday life, without the term. When you log into an app with your Google account, a connection like this is working in the background. The same applies when a learning app writes assignments into your calendar. In companies, connectors typically link programs like Salesforce, SAP, Slack, or Microsoft 365.

In the AI industry, the term has become especially visible since 2024. Major providers allow chatbots to be connected to company data and explicitly advertise their connectors. At the same time, a common standard has established itself: the Model Context Protocol. The idea behind it: not every vendor builds its own connections, but everyone uses the same plug shape.

For investors, this is more than a technical detail. Whoever offers many connectors ties customers to itself, since switching then means a lot of rework. The connector should be distinguished from the platform above it: the platform controls processes and rules, while the connector is just the individual link within it.