Local AI model or API? And the third way
Run a model on your own machine or call it at a provider: what each way gives and costs, what leaves for a third party when you call an API, and the third way, a dedicated machine in Switzerland, by the hour, delivered ready to use.
An AI model is used either locally or through an API. Locally, you control the installation, the versions, the workflows and the storage of your data, at the cost of hardware and maintenance; through an API, you get access to a model your machine could not run, but your images and instructions leave for the provider, under its terms. A workflow can combine the two, project by project. Axiris offers a third way: a dedicated machine in Switzerland, rented by the hour and delivered ready to use, the control of local without buying or administering the workstation, and, in the Nova service, the choice of engine according to the level of confidentiality you set.
Two ways of using a model
Some models run directly on a workstation or a server. Others are only accessible through a remote service, an API. Both have their place in a professional workflow; the question is which one serves what, and where your data goes.
Local execution
Running a model on your machine gives you control of the installation, the files, the versions, the models and the workflows, some of the automation, the storage of your data and the performance. Nothing leaves: your sources, your renders, your brand guidelines stay with you.
This way requires suitable hardware (How much VRAM to generate AI images?) and maintenance: an environment you install, update and troubleshoot. The ComfyUI section explains what that involves.
APIs and remote services
An API lets you use a model running on its provider's infrastructure. You do not need the GPU able to run it, and some models (Gemini Image, GPT Image) are used this way.
In return, you have to account for the pay-per-use cost, the terms of use, the availability of the service, the limits of the API and the integration options.
What leaves for a third party
When you call an API, what you send leaves the machine: the reference images, the visuals to modify, the instructions, and sometimes what they reveal about a product before its launch or a campaign before it goes out. What the provider does with it is written in its terms, which you need to read.
For a company, this is the question, and it is decided project by project, not once and for all: everything stays local, a provider you name, or the cloud allowed when you ask for it. The three levels are explained in Your data does not leave, unless you decide it does.
A workflow can mix the two
There is no need to pick a side. A pipeline can use a local model for some operations and a remote model for others; ComfyUI now integrates some remote services into a single workflow: Models in ComfyUI.
At Axiris, the work decides: local when control and integration justify it, remote when a service brings a genuinely useful capability. We work with the clouds when they serve the project; sovereign is an offer, not a rule imposed on everyone. The objective remains the production pipeline, not the purity of the architecture.
The third way
Between "at my place" and "at a provider", there is a third way: a dedicated machine, in Switzerland, rented by the hour, delivered with the environment ready: the system, ComfyUI, Nova. You have the control of local without buying or administering the workstation; you install your own models and workflows, and your files stay in Switzerland: A sovereign GPU in Switzerland, by the hour.
And in the Nova service, you do not choose the route: you set the confidentiality level, and Axiris chooses the engine for each task, local, Swiss or cloud. The direction is explained in the news item of 6 October 2026: Axiris is preparing its AI router.
Contact request: a first conversation, free and with no commitment. Every practice is different; it is by listening to you that we propose what suits you.