When software calls itself “local-first,” the useful follow-up is a practical one: where does my work happen, and what information is sent elsewhere?
In Upfyn, the desktop is the place where local project files and machine tools are used. The model that helps reason about the task may run through an online provider or through a compatible local setup.
Understanding those two parts makes it easier to choose the right arrangement for your work.
The project is based on your desktop
When Upfyn opens a local file, runs a project tool, or works with an installed browser, that action happens in the desktop environment. You do not need to create a separate hosted project just to use those local resources.
The mode and permissions still matter. Analyst can read the active project and keeps its own file outputs in the output folder. Developer can use the wider project and development tools within the access you choose.
The computer is the working environment, and the task settings define how the assistant may use it.
A cloud model needs the relevant context
If you choose an online model, the provider receives the prompt and context needed for its response. That can include relevant file content, tool results, and conversation material.
For example, a cloud model asked to summarise a local report needs information from that report. The fact that the original file is stored on your computer does not mean every part of the summarisation happens offline.
Review the selected provider and its policies when choosing the route. UpfynAI, a supported provider key, and a supported subscription route have their own arrangements.
A local model changes that part of the workflow
A compatible local model, such as one served through Ollama or LM Studio, can handle the model turn on your computer. This can be useful when you want the reasoning to stay on the machine or need to work without a cloud model connection.
The model still needs to be capable enough for the task, and your hardware affects performance. Try representative work before relying on a particular setup.
Online services remain online services. A local model cannot fetch fresh Gmail messages or research a live website while the computer has no internet connection.
Different apps have different data paths
Research keeps working documents in the active project. Design keeps its projects with the workspace. AI Studio uses a local media library and timeline.
Team is account-backed collaboration, where invited members work with shared information. Share moves selected files across a trusted local network. Operator provides a temporary remote-assistance session.
These distinctions are more useful than assuming every feature stores and transmits information in exactly the same way. Look at the specific action you are taking and the service it uses.
Remote access still reaches your computer
A paired phone or browser lets you reach the desktop. Local work continues on that computer, so it must be running and connected when needed.
The same principle matters for scheduled tasks using local resources. A durable schedule can preserve the job and timing, but it cannot make a powered-off computer read a file.
Ask four practical questions
Before a new kind of task, identify where the sources live, which model receives the context, which tools will act, and where the result will be saved or sent.
Those answers make the setup understandable. You can choose a route that fits the task and keep the permissions aligned with what you intend the assistant to do.
Explore Upfyn Desktop · Read the current security information