Israeli startup Uvi has closed a $7.5 million seed round led by CT Innovations, with Cardumen Capital, Manu VC and 91 Ventures participating, to commercialize a knowledge-graph "reasoning layer" that the company says cuts the cost of an enterprise AI answer from as much as $2.00 to between $0.10 and $0.16.

The company, founded in Tel Aviv in 2023 under the name Naboo by CEO Gilad Salinger and CTO Dror Wolmer, is attacking a structural inefficiency in enterprise AI: every agent an enterprise deploys starts work knowing nothing about the company it serves, and reconstructs that context from scratch — with the enterprise paying a frontier model every time.

What does the reasoning layer actually replace?

Uvi's argument is that RAG (Retrieval-Augmented Generation), the dominant technique for grounding LLMs in enterprise data, works well for static knowledge bases but fails in living R&D environments. A developer's AI assistant does not need "the 10 most similar document chunks," the company says. It needs to know which code is involved, who owns it, what decisions were made, what changed recently, and what the developer is allowed to see.

The core artifact is what Uvi calls a Decision Graph:

• Decisions are modeled as first-class nodes carrying owner, trigger, blockers, dependencies and supporting evidence.
• Live joins and ETL run across GitHub/GitLab, Jira/Linear, Slack, Confluence/Notion, Datadog/Splunk and internal services.
• Agents query it via GraphQL or MCP (Model Context Protocol) and receive a structured chain of relevant decisions rather than a list of document chunks.
• The system runs on-prem or in the customer's virtual private cloud, with native RBAC enforced at retrieval and zero data egress.

Salinger frames the problem as one that evolves with AI maturity. "Enterprise context is a problem that gets bigger as AI adoption grows," he said. "Early use cases exposed accuracy gaps. Today, the conversation has shifted to cost and tokens. But as companies move into more complex agentic workloads, the real bottleneck becomes speed. And speed depends on context."

The positioning is deliberately distinct from Glean-style enterprise search and from standard knowledge graphs: the layer is built around intent, company-specific semantics and actionable reasoning chains — a substrate for agents that act inside organizations rather than summarize documents.

Where are the production numbers?

Uvi runs in production at Lemonade and monday.com, both public companies, and at Melio, now part of Xero. Integrations cover Slack, codebases, coding sessions, meeting summaries, Jira and Confluence, with more in development. Employees see only what their existing access rules allow, and every answer traces back to its source, the company says.

A company benchmark illustrates the economics. Uvi ran an identical set of questions twice — once through its context layer, once through a do-it-yourself agent over the same sources — and measured cost per correct answer: total model spend divided by questions answered right, not merely answered. The DIY agent came out roughly 5x more expensive per correct answer, burning tokens searching for context a structured map would have supplied directly.

The spread widens on open-ended questions where agents get stuck in search loops: there, Uvi claims to be up to 46 times cheaper. Built the way most enterprises build today, an AI answer costs $0.75 to $1.41 and can exceed $2.00; built on Uvi, the company puts it at $0.10 to $0.16.

Uvi offers a monday.com example of why retrieval alone falls short. Asking what shipped last week does not lead to a single record: a feature may exist across a ticket, a pull request and a feature flag that only make sense together. Without modeling those relationships, an agent can return a list of tickets and still fail to answer the question.

Who is behind it, and who else is chasing the same prize?

Salinger previously served as VP technology at Nutrifresh Services, which acquired his earlier company Fetchy. Wolmer is an ex-security researcher and former CTO of Cognigo, acquired by NetApp, and holds three US patents — including US-12468695-B1, issued November 11, 2025, covering "a system and method for querying knowledgebases utilizing a language model based on consolidated identity context."

Uvi is model-agnostic by design: models and agents can be swapped without forcing each new system to relearn how the business works, because the company's context — not the model — is the durable asset. Keeping that semantic layer, knowledge graph and reasoning layer permissioned and in sync with live data is, the company admits, heavy lifting no engineering team wants to own; Uvi does it as a managed service.

The company is not alone in pursuing organizational memory for AI agents. HYCU, Jedify and MinIO are all exploring variations on the same theme, and enterprise context is likely to become contested ground as agentic workloads scale and token economics sharpen.

Uvi will spend the seed capital on building its US team, expanding go-to-market efforts and continued product development.