Around 200 microseconds – that’s how fast Alexander Malandin says his database can find a single record among billions, and it’s the number behind his claim that SereneDB built the world’s fastest search engine.
On a 10-billion-record dataset, roughly what a single enterprise generates in log data over half a day, Malandin says SereneDB outperformed both Elasticsearch and ClickHouse in its benchmark tests, beating ClickHouse’s analytical engine, its main rival, by a factor of 10.
“We first benchmarked it against the implementation of IResearch that we created at the company we worked at previously (ArangoDB), and that still remains there unchanged, and beat it by a few hundred times on performance. Then we tried it against everyone else in search, and it turned out we beat all of them too. So, we took ClickHouse and measured ourselves against it,” says Malandin, co-founder and chief executive of SereneDB, to Tech Funding News.
SereneDB backs the claim with its own open benchmark, called SearchBench, built on the same model as ClickHouse’s own ClickBench, so anyone can try to break the numbers themselves.
Interestingly enough, ClickHouse’s own team was the first to take him up on it, submitting changes SereneDB folded into a retest, a validation Malandin treats as proof the whole exercise worked exactly as intended: “Now we tell everyone, honestly, that we’re the fastest search in the world.”
A glossary rather than a table of contents
SereneDB was founded in Berlin in 2025 by Malandin, Andrey Abramov, and Valery Mironov as an open-source, Postgres-compatible database. Its engine combines vector search, full-text search, and analytics so companies can query fast-moving data without stitching together separate indexes, caches, and dashboards.
The company has released its code under the Apache 2.0 license, making it easy for users and contributors to adopt and build on the technology.
Malandin explains the speed with a simple example: “Think of an old encyclopedia. A table of contents helps you narrow things down, but you still have to flip through many entries to find the answer. A glossary works differently: look up a term once, and it instantly shows every page where it appears. Standard database indexes work more like a table of contents. SereneDB builds more like a glossary.”
That design is what lets SereneDB’s core search library return results in around 200 microseconds, roughly the time it takes light to travel 60 kilometers. It shifts the work to write time so queries can run with almost no delay.
The industry impact is clear: AI systems are only as fast as the infrastructure behind them, and today search and analytics are often split across separate tools. That creates delay, complexity, and cost. For AI agents running at scale, those bottlenecks can mean queues, timeouts, higher bills, and added strain on the systems that connect everything.
On competition, he points out that Elasticsearch, OpenSearch, and the search layers in MongoDB and Neo4j all rely on Lucene, an open-source library dating back to 2001. His view is that people often assume search has fixed limits when, in his words, the real limit is Lucene.
While engines like Lucene, Tantivy, and IResearch all build on the concept of an inverted index, per se the “glossary”, SereneDB uses IResearch as a next-generation engine that resolves many of their inherent limitations. That is how the Berlin-based startup is trying to rethink the core mechanics of search and analytics for the AI age.
Where small startups start looking like enterprises
SereneDB isn’t chasing massive enterprises like Booking.com, Uber, Visa, or JetBrains, as these organisations manage hundreds of Elastic nodes that have virtually no reason to migrate.
Instead, the startup is targeting what Malandin describes as vertical AI: nimble teams leveraging AI agents across vast datasets to perform work that previously required entire departments.
As early-stage firms deploy automated agents to sift through information, their data demands quickly mirror enterprise-scale workloads without the accompanying headcount. “It’s all becoming one continuum,” Malandin observes.
To illustrate the impending surge, he highlights Booking.com, which processes roughly 200,000 trip requests every second. Assign 100 AI agents to each user, he suggests, and query volume could theoretically explode to 20 million per second — a load legacy setups simply cannot support.
Consequently, a lean startup operating enough agents could soon match the query output of an Uber or Booking.com while maintaining a tiny team. In Malandin’s view, the next systemic crisis might not be biological, but a severe infrastructure bottleneck triggered by unbridled demand.
Highlighting the system’s capabilities, he points to a SereneDB software developer and Dota 2 enthusiast who built an entire game analytics platform in just a fortnight using 30 million public Valve match files. The platform pins down the exact turning point in a match, pinpointing errors that dipped a team’s win probability by 5-10 %. A project like that typically takes a dedicated analytics squad months; for a single engineer, it takes two weeks.
Malandin expects this rapid execution model to spread quickly across other data-heavy domains, including biotech and genomic research.
Eight people, one product, a round to match
SereneDB employs a small team of eight people, most of whom are software engineers, and all are based in Berlin. In December 2025, the company raised $2.1 million in a pre-seed financing round, with the Berlin-based fund Entourage and the German investor High-Tech Gründerfonds as lead investors.
The company’s initial commercial success has been achieved through a small number of resale partners obtained by word of mouth. Malandin intends to expand this group into a close network of startups that can support one another before eventually pursuing their individual paths.
The company is preparing to launch a new funding round in early 2027, with the aim of developing two commercial products: a managed hosting service and an on-premises licence for customers unable to use the cloud due to compliance requirements.
As market appetite for ultra-low-latency grows, SereneDB’s core challenge will be to convert raw speed into sustained commercial dominance before competitors narrow the gap.
