When AI agents must reconstruct organizational knowledge by sifting through Slack messages, Jira tickets and pull request histories to answer routine questions, enterprises foot the bill for redundant context assembly. Uvi, an Israeli startup that launched publicly Wednesday, raised $7.5 million in seed funding from CT Innovations, Cardumen, Mana VC and 91 Ventures to address this cost multiplier through a centralized memory architecture.

The company's internal testing found that agents built on frontier models and wired directly to six enterprise systems incurred roughly eight times higher costs than necessary, with the excess driven primarily by repeated context retrieval rather than actual reasoning work.

The price of repeated knowledge discovery

Uvi co-founder and CEO Gilad Salinger frames the problem in stark terms: "Developers are paying frontier prices for amnesia." He notes that while models themselves are rented commodities, organizational context remains proprietary. "Every quarter there's a new model that is faster and cheaper, and every quarter the company's memory of why it built things the way it did stays exactly where it was: in people's heads, in old tickets, in Slack threads nobody will read again."

Uvi's approach extracts that institutional memory from dispersed locations—people, tools, chat logs—and consolidates it into a single accessible layer. Agents can then reference this unified context regardless of which model powers them. If a company switches to a newer frontier model tomorrow, no relearning cycle occurs. Salinger emphasizes that Uvi ensures this memory remains actionable rather than merely archived.

Context is unique to every company and demands a deep understanding of how that company defines its work, from what counts as a feature to how two departments communicate.

Gilad Salinger

Testing methodology and performance claims

Uvi's architecture layers a semantic knowledge graph and reasoning engine atop synchronized, permissioned live data from enterprise systems. The company tested its platform against more than 900 actual employee questions using identical frontier models on both configurations. These tests remain internal and have not undergone independent verification.

The first test configuration ran an agent directly connected to six enterprise systems—Slack, Jira, GitHub and others. Each query forced the agent to traverse these systems, assemble fragmented information, and map connections before responding. This approach required 15 to 35 round trips per question.

The second configuration deployed the same frontier model with Uvi's platform. Because Uvi had already mapped relationships between company decisions, conversations, code and processes, the agent retrieved its context in a single step. Salinger describes the contrast: "The first agent was doing detective work every time someone asked a question. The second already had the case file. And here's the ridiculous part: The first approach cost roughly eight times more and took more than twice as long. We weren't comparing a stupid model against a smart one. We were comparing the same brain with and without an understanding of the business it was supposed to work for."

Cost differential in production scenarios

Uvi quantifies the expense gap across typical query patterns. Built using conventional direct-connection approaches, a single answer costs between $0.75 and $1.41, potentially exceeding $2.00. The same answer via Uvi's platform costs $0.10 to $0.16. On open-ended queries where agents enter search loops, Uvi claims cost reductions reaching 46x.

Salinger argues the deeper concern transcends mere waste. "The real worst case is not waste. It is losing the race." He contends that enterprises competing in markets subject to rapid disruption depend on proprietary information as their durable competitive moat against AI labs. If companies cannot mobilize that information through agents quickly, safely and with precision, their advantage remains theoretical rather than operational.

Industry awareness and scaling challenges

Frontier model providers understand the centrality of context to agent utility, Salinger suggests. Their investment in enterprise consulting divisions signals this awareness. Yet the economics of context differ fundamentally from model economics. "Context is the opposite. It is unique to every company and demands a deep understanding of how that company defines its work, from what counts as a feature to how two departments communicate. That does not scale the way a model does."

Uvi's dual-layer approach simultaneously interprets information and traces connections between data points. A query about a billing exception retrieves the decision, the stakeholders and the dependencies—not every document mentioning billing. Because Uvi models each organization's specific workflows, identical questions receive different answers across different companies.

The first agent was doing detective work every time someone asked a question. The second already had the case file.

Gilad Salinger

This capability allows the platform to deliver the "smallest correct slice" of context at enterprise-grade speed, security and precision. Salinger acknowledges diminishing returns exist at this optimization level, so Uvi's role involves identifying the sufficiency threshold for each question within each organization.

Integrations and current deployments

Organizational knowledge rarely exists in a single canonical location. A decision might originate in a Slack conversation, become a Jira ticket, evolve through pull request comments, and eventually hide behind a feature flag, with no individual system holding the complete narrative. Uvi reconstructs these distributed stories by mapping connections and sequencing what was decided, why changes occurred, and what consequences followed.

The platform then exposes this reconstructed context to any models or agents the company deploys. Without access to this history, agents risk confidently recommending migrations or architectural shifts that the company rejected years earlier for identical reasons.

The real worst case is not waste. It is losing the race

Gilad Salinger

Uvi reports production deployments at Lemonade and monday.com, both publicly traded companies, as well as Melio, now owned by Xero. The platform currently integrates with Slack, code repositories, coding session logs, meeting transcripts, Jira and Confluence, with additional connectors under development.