xmemory, a London-based developer of a memory layer for AI workflows, has raised $4 million in pre-seed funding as it emerges from stealth and begins early deployments with select partners.
The company is building infrastructure designed to improve how AI systems store, structure, and retrieve information. Its platform focuses on optimizing the “write path,” organizing and cleaning data upfront so that downstream queries can be answered with greater precision and reliability. This approach differs from traditional retrieval-augmented generation systems, which primarily interpret context at read time.
According to the company, the goal is to enable AI systems to accurately answer the majority of critical business and system questions, while leaving only a small portion of queries to broader semantic search techniques.
The funding round was led by Fly Ventures, Begin Capital, AAL VC, 33East, and Inovia Capital, with participation from angel investors and industry figures including Vadim Barshtak, Mark Shmulevich, Alexey Dosovitskiy, and Konstantin Vinogradov.
The company is currently operating in closed beta and is launching pilot programs with design partners while selectively onboarding additional companies to help shape the product.
xmemory was founded by Alex Petrov, who serves as CEO. In a LinkedIn announcement, he confirmed the company is coming out of stealth and highlighted its focus on building a high-precision memory layer for AI systems, emphasizing reliability, observability, and governance as core design principles.
KEY QUOTE:
“We’re coming out of stealth with xmemory – the memory layer for AI workflows, built for precision, reliability, observability, and governance. We’re currently in closed beta, launching pilots with existing design partners and inviting a small number of new companies to work with us in shaping the best memory solution in AI. Working on one of the most interesting and technically complex problems in one of the fastest-moving markets in history is a real privilege – and a genuine engineering joy. I’m deeply grateful to everyone backing us on this journey, especially our $4M pre-seed investors. Our approach to memory is different. Rather than trying to interpret endless raw or compacted context at read time, we focus on optimizing the write path for specific memory records – structuring, cleaning, and organizing data upfront so reads become faster, more reliable, better aligned with the questions you already know matter most. In short, we believe 90% accuracy for generic contextual memory is not enough. A better approach is to answer 95% of the most important business and system questions with precision – the ones you know you’ll ask – and leave the remaining 5% to semantic search or RAG.”
Alex Petrov, Founder & CEO of xmemory
