When an answer needs further steps
Imagine comparing several documents and preparing a summary of their differences. You need to find relevant passages, read them in context and separate what is written from what you are inferring. Velia is an agentic assistant for macOS: the model can choose from the available tools, observe the result and decide on the next step.
In the prototype, conversation works alongside document reading, research, persistent notes and tools for the Mac. The capabilities you can actually use depend on the permissions granted, the components available and the connection to the chosen model.
A cycle that makes the work observable
- Clarify the goal
- Choose a step
- Use a tool
- Read the result
- Continue or stop
When comparing documents, the work might begin by listing the files, continue by reading relevant sections and end with a summary that identifies its sources. This illustrates the method: assessing a delivery requires the documents actually read and the differences actually found.
Memory, continuity and control
Persistent notes help recover goals, constraints and decisions after an interruption. Their sources matter: a recorded assumption remains an assumption until checked. Access permissions determine what the assistant can read or change; confirmations depend on the operation and on the permissions already granted.
Where the project stands
Velia is a working macOS prototype, developed in increments. These pages introduce its progress; they do not yet offer a public installation package. The application keeps the work locally and uses external services for model requests, so some activities require Internet access. New capabilities will be described alongside tests showing how they work and where their limits lie.