As enterprises accelerate the deployment of autonomous AI agents, many organizations are discovering that their greatest obstacle is not the model itself, but the lack of institutional context. Tel Aviv-based startup Uvi, formerly known as Naboo, has emerged from stealth with a solution designed to provide AI agents with a persistent, ownership-based understanding of company-specific workflows, decisions, and data relationships. By creating a “Reasoning Layer” that functions independently of the underlying frontier models, Uvi aims to reduce the high costs and accuracy gaps associated with traditional Retrieval-Augmented Generation (RAG) approaches.
The fundamental challenge, according to Uvi co-founder and CEO Gilad Salinger, is that modern AI agents often treat every task as a blank slate. When an agent is tasked with a complex operation, it frequently relies on standard RAG processes, which scan document chunks for semantic similarity. While effective for simple information retrieval, this method fails in dynamic enterprise environments where intent is scattered across disparate systems such as GitHub repositories, Jira tickets, Slack conversations, and CI/CD pipelines. When an agent is forced to reconstruct this missing context from scratch, it leads to inefficient token consumption, slow performance, and, frequently, inaccurate outputs.
Uvi addresses this by constructing a “Decision Graph.” Unlike traditional knowledge graphs that focus solely on data points, Uvi’s architecture models decisions as first-class nodes. Each node incorporates essential metadata, including owners, triggers, blockers, and dependencies. By performing live joins and ETL processes across an organization’s internal toolset, the platform provides agents with a structured chain of reasoning. When an agent queries the system via GraphQL or the Model Context Protocol (MCP), it receives a coherent narrative of why a decision was made, rather than a collection of disconnected document snippets.
The platform is designed for security-conscious, high-performance environments. Uvi can be deployed on-premises or within a customer’s virtual private cloud (VPC), ensuring that data remains local and no egress is required. Native role-based access control (RBAC) is enforced at the retrieval stage, ensuring that agents only access information for which the user has authorized permissions. This architecture has already seen production deployment at major public companies, including Lemonade and monday.com, where it has been used to navigate complex interdependencies between feature flags, code commits, and project management tasks.
Beyond improving accuracy, the economic benefits of a dedicated reasoning layer are significant. Internal benchmarks conducted by the company indicate that standard “do-it-yourself” agentic workflows can cost upwards of $2.00 per correct answer due to model “search loops” and inefficient token usage. By providing a direct map of organizational knowledge, Uvi claims to reduce these costs by as much as 90 percent in some scenarios. Furthermore, because Uvi is model-agnostic, enterprises can swap out underlying LLMs—such as moving between Claude, Gemini, or GPT—without the need to re-index or re-train their organizational context.
With a fresh $7.5 million seed round led by CT Innovations, Uvi plans to scale its U.S. operations and expand its integration ecosystem. As the industry shifts from simple document summarization to agentic workloads that require active participation in business processes, Uvi’s approach suggests that the future of enterprise AI may lie in the ability to maintain a persistent, searchable, and structured memory of the business itself.
