The challenge
A useful agent inside UpKeep needs to do more than answer a question. It has to navigate a mature application accurately, retain the context that matters, and give users a clear experience across web and mobile.
Our work
Give the agent the right tools at the right time
We designed progressive disclosure and MCP access so Nova can work with the relevant capabilities as a task develops. That includes navigating hundreds of API endpoints and database structures built over a decade.
Manage memory and long-running context
We worked on memory and compaction: retaining useful information while reducing the amount of prior conversation the model must carry. Accuracy and continuity shape how the agent handles context.
Build the customer experience
We developed the agent experience across web and mobile and launched a custom app builder inside Nova, allowing customers to tailor their UpKeep experience to their needs.
Improve the economics of each task
We worked with frontier inference providers and communities to lower Nova’s operating costs, reducing cost per task by more than 80% in early 2026.
The result
Nova brings AI assistance and custom app creation into the product customers already use. We continue to develop its architecture, memory, tool access, and user experience.