Agentic artificial intelligence startup Uvi said today it has raised $7.5 million in seed funding to tackle an expensive problem around business context that’s draining millions of dollars annually from the coffers of enterprises using autonomous AI agents.
It’s backed by CT Innovations, which led the round, alongside Cardumen, Mana VC and 91 Ventures, which joined as participating investors.
Each time a new AI agent is deployed, it knows nothing at all about the company that deployed it. Those agents have to learn everything about the company’s business processes and operations from scratch in a process that’s repeated again and again, racking up significant costs in terms of computing resources.
Business context is crucial to agentic operations. If an AI agent is going to be able to work autonomously for an organization without constant human oversight, it needs to be well versed in the way that company operates.
That means knowledge of its operations, its historical decisions and so on. Until now, the only way agents could obtain this context was to reconstruct it from scratch with each new deployment, and that in turn means paying the model provider for each token used in that relearning process.
Uvi, officially known as Naboo Research Ltd., says enterprises are losing a fortune to this inefficiency. Data from its own benchmark tests shows that internally built AI agents cost the average company about eight times more than they should because of the need to repeatedly gather all of this context before they can get started at work. It’s disadvantageous in terms of speed, too, said Uvi co-founder and Chief Executive Gilad Salinger.
Salinger told SiliconANGLE that the problem is getting bigger the more AI adoption increases. He stressed that in the early days of AI agents, most companies were worried about the accuracy of their responses and the security of the actions they take. But now the conversation has turned towards cost, tokens and speed, where agents continue to struggle.
“The reason is structural, not about intelligence,” Salinger said. “Big companies run dozens of different tools with completely different data shapes, and the ability to pull two slices of data in parallel, say one from Snowflake and another from GitHub, and then reason over that data does not exist for them. So even when an agent has already completed a task, it would fail the same task the next time, because it never retained how the work actually gets done.”
Context is everything
Uvi believes it has hit upon a solution, developiing what Salinger calls a “context reasoning layer” that each company owns. Instead of having each new agent relearn the way the business operates, they can simply tap into this unified institutional memory that’s made available to every AI tool within its information technology environment.
Uvi supplies this context by linking fragmented enterprise systems such as Slack, Jira, Confluence, Salesforce and so on with other data from code repositories and the minutes of internal meetings. It then builds a map that shows how all of this information is linked, building a full picture of why each decision was made, what the decision was and its impact on the company.
This context reasoning layer is built atop knowledge graph that functions as an infrastructure layer, sitting underneath every AI agent that’s deployed by a company. It uses entity mapping to understand the organization’s engineering stack, track its dependencies and how different assets within it relate to one another.
Uvi also analyzes historical context, such as a decision that was made in Slack that was later reversed inside a comment in a document, to reconstruct agent’s reasoning processes. In addition, Uvi has developed a custom semantic layer for the knowledge graph to help AI agents learn the specific company jargon they’re going to come across, ensuring they can interpret terms accurately.
More importantly, Uvi’s context reasoning layer is secure by design. Knowing that knowledge graphs often struggle when it comes to things like user permissions, the company built an access control model that controls access at the application and administration layers, and also uses data tagging.
“We tag data so that a person who leaves a tool, or leaves the company, cannot be surfaced by agents down the road,” Salinger explained. “This is the part most RAG pipelines get wrong. They ingest once and forget who was allowed to see what. Our graph reconstructs relationships across platforms and carries the permissions with it, so there’s no need for a separate security retrofit.
Uvi’s benchmarks reveal huge cost benefits for companies willing to adopt its context reasoning layer. Agents that run without it can cost between $0.75 and $1.41 per response, and in some cases can even exceed $2, Salinger said. But when those agents route their responses through Uvi, the costs fall dramatically, to an average of just 10 to 16 cents per answer.
The cost savings are even more dramatic with open-ended questions, where agents often get themselves stuck in a recurring search loop, relearning the same bit of context again and again to try and respond to a single prompt. In such scenarios, Uvi agents cost 46 times less than their non-Uvi counterparts, while answering questions twice as fast.
Uvi has already convinced Monday.com Inc., Lemonade Inc. and Melio of the merits of its platform. In the case of Monday.com, Uvi helps its agents understand the relationships among Jira tickets, pull requests and feature flags. Without this contextual understanding, agents often retrieve a disjointed list of tickets, Salinger said, forcing workers to manually sort through them to find the answers they need. By tracing the relationships of those tickets, Uvi can dig up the answers much faster.
This is why CT Innovation’s Gwen Cheni is convinced that Uvi is onto something good. She said virtually every company she talks to about AI agents has struggled to build context at scale. “Most are still very far away,” she added. “Uvi is the finished version of a project they have already paid for multiple times over.”
Salinger said he believes that every AI agent will eventually need a context layer to tap into, because there’s simply no other way to do it without sacrificing efficiency.
“In three years, the enterprise AI stack will not be organized around chat interfaces or document search,” he said. “It will be organized around a persistent context layer that every agent, every model and every identity plugs into. Companies that have this content will ship new agents the way they ship with continuous integration/continuous development pipelines today, while those that don’t will be forced to keep paying their models to relearn their business every morning.”