Israeli startup Uvi states enterprises pay leading AI providers to repeatedly re-learn internal business knowledge that exists but remains inaccessible, and its knowledge graph technology can unlock this hidden context.
Uvi notes every enterprise AI agent starts without any understanding of its host organization. Agents must rebuild business context from scratch, with firms incurring model charges each time. Its knowledge graph supplies Claude, Gemini, GPT and other AI tools with missing context behind business decisions, discussions and code. This creates a reusable enterprise context reasoning layer owned by the organization and available to all deployed agents.
Gilad Salinger, CEO and co-founder, commented: “Enterprise context is a growing challenge as AI adoption expands. Early deployments exposed accuracy gaps. Now conversations center on token consumption and cost. But as enterprises adopt complex agentic workloads, speed emerges as the real bottleneck. Speed relies on context: grasping organizational knowledge, parallel changes made by other agents, and decision histories, then filtering only relevant details for each agent to keep tasks progressing.”
Maintaining an up-to-date, permission-aware semantic layer, knowledge graph and reasoning layer synced with live data requires heavy ongoing work most engineering teams prefer to avoid. Uvi delivers this capability as a managed service.
Salinger and CTO Dror Wolmer, based in Tel Aviv, founded Uvi in 2023 under the original name Naboo, to create the foundational layer enterprise AI agents require to operate within organizations, beyond simple document summarization. Salinger previously served as VP of Technology at Nutrifresh Services, which acquired his startup Fetchy. Wolmer worked as a security researcher and CTO at Cognigo, later acquired by NetApp.
Naboo originally developed a “Reasoning Layer”, also called enterprise context or semantic layer, so AI agents can act within businesses rather than merely summarize documents. Enterprise AI agents often suffer poor performance or high costs because institutional knowledge spreads across code, tickets, PRs, Slack threads and documentation, while the rationale behind decisions is rarely formally recorded. Uvi solves this by building a Decision Graph.
Within the graph, decisions are defined as core nodes with assigned owners, triggers, blockers, dependencies and supporting evidence. The platform runs live joins and ETL across common tools including GitHub/GitLab, Jira/Linear, Slack, Confluence/Notion, Datadog/Splunk and internal services. Agents query the graph via GraphQL or MCP (Model Context Protocol) and receive structured decision chains instead of raw document chunks. It can be deployed on-premises or in the customer’s VPC, with native RBAC enforced at retrieval and zero data exfiltration.
The solution goes beyond RAG, enterprise search such as Glean, or conventional knowledge graphs, focusing on intent, business-specific semantics and actionable reasoning chains.
Uvi acknowledges RAG was a major advance granting LLMs access to external knowledge by embedding documents into vectors and retrieving semantically matching text chunks for queries, and works well for use cases like customer support knowledge-base bots.
But enterprise R&D environments differ greatly from static knowledge bases. They are dynamic ecosystems containing code repositories, issue trackers, pull requests, Slack conversations, monitoring dashboards, CI/CD pipelines and documentation, all interconnected, constantly changing and governed by strict access controls.
When an AI agent assists developers, it does not need the top ten similar text snippets. It needs task-specific context: related code, owners, prior decisions, recent modifications, and the user’s data access permissions. Standard RAG cannot deliver this, but the Decision Graph-powered Reasoning Layer can.
Uvi runs in production at public firms Lemonade and monday.com, plus Melio, now part of Xero. It integrates Slack, code repositories, coding sessions, meeting summaries, Jira and Confluence, with further connectors under development. Employees only view data matching their existing access rights, and every response links back to source materials.
At monday.com, querying what shipped in the prior week cannot be answered from one single record. A feature may span tickets, pull requests and feature flags, which only make sense together. Without mapping these relationships, AI agents can return ticket lists yet fail to fully answer the business question.
Uvi ran benchmark tests on queries such as: "Do we have a guide for creating a GitHub app?". Two setups were compared: its context layer and a self-built agent using identical source data, measuring cost per correct answer as total model spend divided by correctly resolved questions. The self-built agent was roughly five times more expensive for each valid answer, wasting tokens searching for information the pre-built knowledge map could supply directly.
Most enterprise AI implementations today generate answers costing $0.75–$1.41, sometimes exceeding $2.00. Built on Uvi, identical answers cost only $0.10–$0.16. For open-ended queries where agents enter repetitive search loops, Uvi delivers up to 46x cost savings.
Uvi maintains model-agnostic architecture: AI models may change, but organizational business context remains consistent. Enterprises can swap models and agents without forcing each new system to re-learn internal business logic.
The company secured a $7.5 million seed funding round led by CT Innovations, with backing from Cardumen Capital, Manu VC and 91 Ventures. Funds will support U.S. team expansion, go-to-market growth and ongoing product development.
Beijing Qianxing Jietong Technology Co., Ltd.
Sandy Yang/Global Strategy Director
WhatsApp / WeChat: +86 13426366826
Email: yangyd@qianxingdata.com
Website: www.qianxingdata.com/www.storagesserver.com
Business Focus:
ICT Product Distribution/System Integration & Services/Infrastructure Solutions
With 20+ years of IT distribution experience, we partner with leading global brands to deliver reliable products and professional services.
“Using Technology to Build an Intelligent World”Your Trusted ICT Product Service Provider!
Twoja wiadomość musi mieć od 20 do 3000 znaków!
Polish
