Local or cloud?
The right boundary is set by the use.

When local is worth considering

Running locally can meet an offline need, a requirement to control the hardware, or a regular task that an available model covers well. In return, it requires sizing, updates, access protection and ongoing operation.

A server you have bought does not become free to use: energy, administration, storage and replacement still have to be planned for. Its capacity must be tested with the expected number of users and documents.

What a cloud API can offer

External capacity can avoid an upfront purchase or handle a difficult task that existing hardware covers poorly. You need to examine availability, cost, the data sent, the service terms and what happens if the service is interrupted. The application remains responsible for the user journey around that call.

Example of a hybrid architecture

In a fictional document-preparation scenario, the original files stay on the company’s infrastructure. A first local step selects the useful extracts. Only the items authorized for processing are sent to an external service, and the output then returns to the working system.

This arrangement is not automatically confidential: the extracts may themselves contain sensitive information. The selection, data minimization and permissions must be verified. If nothing is allowed to leave, test a fully local route or reduce the need.

A short decision grid

• Which data can be processed where?
• What quality does this task need?
• How many simultaneous requests must be handled?
• What waiting time is acceptable?
• Who maintains the system, and how does it work during an outage?

The answer may differ from one task to another. Saphan can compare the routes on the same cases before choosing an architecture. See also the Ollama documentation to tell its local runs from its cloud features; the name of the software alone does not settle the question.