The models we use and how your data is processed
This article explains the AI technology behind the Command Center — what it is, who provides it, and exactly what happens to your data when you use it.
The model powering the Command Center
The Command Center is built on GPT models from OpenAI. These are the same models that power a range of widely-used AI products — they're capable, well-tested, and operate under strict data handling agreements.
Paystack built a custom layer on top of these models — called canvas-api — which handles how your data is retrieved, what gets passed to the model, and how responses are returned to you. The model itself generates the language; everything around it is built and controlled by Paystack.
How a query is processed
When you type a question in the Command Center, here's what happens:
project-canvas-apireceives your query and identifies what data is needed to answer it.A trimmed version of the relevant data is retrieved from your account — not your full dataset, only what's pertinent to the question.
Your query and that data are sent to OpenAI's model for processing.
The model generates a response, which is passed back through canvas-api and displayed to you.
The whole exchange happens in seconds. If your connection drops mid-response, the session can resume — you won't lose the conversation.
What OpenAI does (and doesn't) do with your data
OpenAI processes your data to generate a response. That's it. Under our agreement with OpenAI:
Your data is not stored by OpenAI after processing.
Your data is not used to train any OpenAI model.
Your data is not shared with any third party.
Paystack does not use your data to train AI models either — now or in the future — without explicit consent.
Validation and hallucination prevention
One of the risks with AI models is that they can sometimes generate plausible-sounding answers that aren't grounded in real data. We've built against this deliberately.
project-canvas-api uses a deterministic harness with strict validation — which means every response the model generates is checked against your actual account data before it's shown to you. The AI cannot invent figures. If it can't find a reliable answer in your data, it will say so rather than guess.
Why we don't show the AI's reasoning
When the Command Center gives you an answer, it doesn't show the steps it took to get there. This was a deliberate choice. Displaying the model's reasoning process would expose details about how our systems work internally — details we consider proprietary. The answer you see is accurate and grounded in your data; the reasoning behind it stays within our infrastructure.