The v/h journal / Document
What we mean by an agent under the hood
The complicated decisions should happen inside the product.
For v/h, an agent under the hood means a local model connected to the device's inputs and tools. It can interpret the task, act through those tools, and check what happened.
The owner should have a clear way to start the job and understand the result. They shouldn't need to choose a model, write a prompt, or troubleshoot the agent's reasoning.
The agent also has to justify the cost and complexity it adds. We would look for a better result, less effort from the owner, or the ability to handle ordinary variation that a simpler mechanism cannot.
The document below explains the product principle we're building around.
What would a device need to show you before you'd trust it to finish a task?

Keep reading
Easy to use and useful to many people


