Local, cloud, or somewhere in between
Where the agent runs, where the model runs, and what each choice actually means for your data, your costs and your results.
These are two separate questions
People collapse "where does the agent run" and "where does the model run" into one decision, and they are not the same. The agent is the part that holds your documents, your rules and the record of what it did. The model is the part that reasons. They can live in different places, and for most businesses they should.
Agent local, model over an API
The common arrangement. A computer in your office holds everything about your business: your files, your history and the record of what the agent did. When the agent needs to think, it sends a request to a hosted AI model and gets an answer back. That request carries whatever the task needs, which can include the contents of your files and customer names and contact details, so those details do reach the AI provider. Your files as a whole stay in your office. This is cheaper, faster and markedly more capable than running everything locally.
Everything local
The models run on your hardware too, so nothing goes out at all. This is the right answer for some work, and it costs real money: capable models need far more compute, and they are still slower and less able than the hosted ones. Anyone who tells you a small, cheap computer runs everything locally with no trade-off is selling you something.
How to choose without agonising
Sort your workflows, not your business. Most businesses find that one or two processes genuinely cannot have their content leave, and the rest are fine over an API. Put the sensitive ones on the strict lane and stop paying the local-model tax on everything else.
If you want this applied to your own shop rather than read in the abstract, send us the workflow and we will write back.