“Developers are paying frontier prices for amnesia”: Uvi wants to stop the agent relearning brain drain
The New Stack Adrian Bridgwater ● Covered by 2 sources
Uvi says AI agents are wasting money by relearning company knowledge over and over. Its pitch: give them one shared memory layer, and the same answer can cost far less.
Based on reporting by The New Stack, Adrian Bridgwater — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Every time an AI agent has to comb through Slack, Jira, and old pull requests to answer something a company already knows, the customer pays for the same search again. Uvi, an Israeli startup that came out of stealth on Wednesday, thinks that hidden repetition is one reason frontier-model agents get so expensive in practice. The company just raised $7.5 million in seed funding from investors including CT Innovations, Cardumen, Mana VC, and 91 Ventures.
Gilad Salinger, Uvi’s co-founder and CEO, puts the problem bluntly: the model may be rented, but the context belongs to the company. His argument is that every new model release may get faster and cheaper, yet the memory of why a business built things a certain way still sits in people’s heads, old tickets, and Slack threads. Uvi’s job, as he describes it, is to pull that memory out of scattered tools and make it available in one place that any agent, on any model, can read.
The company’s architecture leans on a semantic layer, a knowledge graph, and a reasoning layer sitting on top of live, permissioned, synchronized data. Uvi says it tested that setup against more than 900 real employee questions using the same frontier model on both sides. On one side was an agent connected directly to six systems, including Slack, Jira, and GitHub. It had to make 15 to 35 round trips per question. On the other side, the agent got the context in one step through Uvi.
The claimed result is not subtle. Uvi says the direct approach cost roughly eight times more and took more than twice as long. It also says answers that would cost $0.75 to $1.41 in the usual setup can fall to $0.10 to $0.16 with Uvi, and that some open-ended questions can be up to 46 times cheaper when agents get stuck in search loops. Those tests are internal and not independently verified, but the point is clear: the expensive part may not be the model’s reasoning so much as the model’s amnesia.
Uvi says its software is already running in production at Lemonade, monday.com, and Melio, now part of Xero. It connects to Slack, codebases, coding sessions, meeting notes, Jira, and Confluence, with more integrations on the way. The startup’s bigger bet is that companies do not just need smarter models; they need a way to stop paying frontier prices for information they already paid to create.
My take — AI-written commentary, not fact-checked reporting
This is exactly the kind of unglamorous infrastructure pitch that usually turns out to matter more than another shiny model benchmark. Open or closed, the real bottleneck is still company memory, and a lot of AI hype quietly assumes knowledge will just magically assemble itself. It won’t, which is inconvenient for everyone except startups selling the map.
Read more about this at: The New Stack