Most AI products begin by asking you to move the work into their world. Upfyn begins on your computer, where the files, browser sessions, tools and projects already live.
That distinction defines the product. The desktop application is the execution engine. A model can reason about a task, but Upfyn is the layer that supplies the right context, applies permissions and carries approved actions into the machine.
Upfyn is not a coding agent. Voice is the natural front door to a broader assistant: talk about an analysis, a document, an app, a repeated operation or a development task, and Upfyn coordinates the parts of the machine that outcome requires. Coding belongs inside Developer mode; it does not define the whole product.
One assistant, two modes
Analyst is the simpler workspace for research, documents, connected apps and recurring work. It can read the active project and keeps its own file changes inside the project’s output folder.
Developer lets the same conversation coordinate project files, terminal, browser, Git, review tools, goals, side chats and specialist development work.
They are not separate products. Both modes share the composer, voice, model choices, skills and permission system. You choose a mode based on the work in front of you.
The model is not the product
You can use UpfynAI, a supported provider key, a local model, or a supported CLI subscription. Changing the model changes the reasoning resource; it does not replace the workflow, permissions or desktop tools around it.
Under the surface, Upfyn separates orchestration, planning, context, execution, verification and memory. You still work with one assistant. The separation exists so a long task does not depend on one model remembering every responsibility at once.
Start with an outcome
Do not begin by configuring everything. Pick a real folder and state one outcome: understand these reports, draft a document from these sources, inspect this project, or prepare the work that repeats every Monday.
Upfyn should make the movement between tools disappear. That is the useful test—not how many features you can switch on.
What local-first means here
Local-first does not mean every model must run offline. It means the desktop remains the place where project files are opened, tools execute and permissions are enforced. If you choose a cloud model, that provider receives the context needed for the reasoning turn. If you choose a compatible local model through Ollama or LM Studio, the model turn can remain on the machine as well.
This distinction matters because “local” is often used as a vague privacy slogan. In Upfyn it describes an architecture and a boundary. Remote access from a browser or phone reaches the desktop through a relay. It does not quietly create a hosted copy of the project and continue in a cloud worker.
The practical consequence is simple: the desktop needs to be available when a remote session or scheduled job requires local resources. A powered-off computer cannot execute local work. Upfyn states that limit because a trustworthy assistant should make its dependencies visible.
Tools join the conversation
The assistant is useful because it can work across more than a text box. It can use project files, a signed-in browser, connected services and focused desktop apps. Each capability enters through the same permission system rather than becoming a separate assistant you have to brief again.
When you connect an app through its OAuth flow, the authorized tools should become available to the current chat as soon as the connection completes. You should not have to start a new conversation and reconstruct the goal. The conversation is the unit of work; the connector is a tool added to it.
The same principle applies to skills. A skill is a reusable working method attached to a request. The user message can show that skill as part of the instruction, while the assistant still reads the entire request as one coherent outcome.
Focused apps, not a crowded super-app
Upfyn currently includes six focused apps. Artifact turns research and conversation into an editable document. Design covers graphics, presentations, whiteboards and multi-page websites. Share moves selected files directly between trusted devices on a local network. Team keeps channels, tasks, documents, calendars, files and calls together. Operator provides consent-gated remote assistance. AI Studio provides a project-local media library and timeline for image, video and audio work.
These apps do not compete to become the home screen. Use one when the work needs its surface, then return to the same project and assistant. Voice and computer control are shared capabilities in the composer, while recurring work lives in Analyst. They are not extra app shells to install or market as separate products.
Authority is part of the experience
An assistant that can act on a computer needs clearer boundaries than a chatbot. Analyst deliberately narrows writes to output. Developer can use the broader project and development tools, but still follows the active permissions. Browser actions, external tools and remote control expose their own approvals where the consequence requires them.
This is why a good Upfyn instruction names both the outcome and the boundary. “Prepare a supplier exceptions report from this folder, write it under output, and do not send anything” is better than “handle these invoices.” It tells the assistant what success looks like and where human authority begins.
You can widen the boundary later. Starting narrow makes the first result easier to inspect and teaches both you and the assistant how the work should be shaped.
A useful first hour
Choose a folder whose contents you understand. Open Analyst and ask for an inventory or summary with no external actions. Review what it read and where it wrote. Then try Artifact if the result should become a document, or schedule the job if the same input appears regularly.
When you need terminal, Git, project-wide changes or deeper inspection, open the same kind of work in Developer. The shift should feel like adding the machinery required by the task, not changing assistants.
The first hour is successful when you understand three things: what the assistant can see, what it may change and which result you can verify. Everything else—models, apps, connectors and specialist responsibilities—can be added when the work calls for it.
Ask Upfyn
Open this post from the desktop notification and ask:
Based on this guide and the project I have open, should I begin in Analyst or Developer? Explain the boundary and propose one safe first task.