The first game Alexey Milovidov designed had no computer to run on. It lived on paper, in mazes and imagined screens drawn on photocopied sheets his father brought home from work. The family machine belonged, in the natural order of older siblings, to his two brothers. Alexey watched them play. Then he nagged one of them to turn his sketches into software. “I became a manager much earlier than a programmer,” he later joked.
The joke improves with age. Milovidov is now co-founder and chief technology officer of ClickHouse, the company built around the analytical database he created. Yet the childhood arrangement still feels familiar: invent a thing, persuade somebody to help, and refuse to accept that the machine cannot do it. His brother eventually showed him BASIC. A command produced a circle on the screen. To a child not yet ten, it felt like a magic wand.
There is no magic in ClickHouse, although its speed can tempt the adjective. There is a much more useful story: one of pressure, observation and thousands of small decisions. The database began because web traffic would not politely stop growing while its engineers considered their options.
Five minutes, or the queue begins
Milovidov joined Yandex in 2008 and landed on a new service with the internal name Convertometer. It launched as Yandex.Metrica, a web-analytics product receiving an unruly stream of page views, clicks and visits. MySQL held pre-aggregated reports. C++ programs handled what MySQL could not. The volume kept rising. New data arrived in real time, and every slow interval became debt for the next.
He searched promiscuously. Databases, libraries, compression methods, unfamiliar data structures mentioned over lunch: anything that might survive production was invited to audition. The first answers were two prototypes. OLAPServer stored non-aggregated data by column. Metrage used background merges for continuously updated reports. Each solved a portion of the problem. Neither was the end.
The interesting leap was to combine their virtues: columns for analytical speed, a merge tree for real-time updates and locality, then a genuine query language and general data types. This was the beginning of ClickHouse. It did not descend from an established database. Milovidov started the machinery from scratch, which is both an engineer’s liberty and an engineer’s excellent way to lose several years.
The first preserved commit, dated May 29, 2009, replaced slow time functions from the standard C library. They had appeared in the profiler and annoyed him. It is a splendidly unceremonious opening: no manifesto, only an irritation with measurable consequences.
A house assembled from useful rooms
By March 2012 there was a ClickHouse server; later that month, a client. The system could be deployed for logs and queried with SQL. MergeTree followed, sorting arriving data in the background so that requests by website could run quickly. In 2014, replicated storage helped put customer-facing queries into production for Yandex.Metrica 2.0.
From internal experiment to independent company
The architecture grew under unusually candid conditions. Yandex infrastructure teams periodically switched off an entire data center as a drill. Services that had treated redundancy as a future refinement received a memorable lesson in the present tense. ClickHouse acquired replication across data centers and automatic repair. Production, in Milovidov’s telling, is less a destination than a disagreeable colleague who is frequently right.
He wanted code to be read, too. In interviews he has argued for detailed comments and programs that can be followed like a textbook or blog post. Complexity may be necessary; obscurity is merely expensive. This preference appeared outside the code. When ClickHouse first needed internal users, Milovidov made its documentation into a coherent site that could be read from beginning to end. Adoption begins when somebody else can enter the room without asking where the light switch is.
The name itself is compact documentation. ClickHouse means Clickstream Data Warehouse: the data it first understood and the category it intended to remake. By late 2014, teams across Yandex were using it for advertising, commerce, infrastructure and business analytics. CERN had deployed it for work connected to the LHCb experiment. The house had one great defect. Almost everyone who might need it still lived outside.
The frightening alternative
At conferences and in technical blogs, Milovidov saw other engineers recreating familiar systems because the available analytical databases did not fit their workloads. He reached an uncomfortable conclusion: if he did not make ClickHouse available, somebody else would eventually fill the gap. He described that prospect as “really scary.”
Open sourcing a production database requires more than pushing a repository into daylight. He prepared the argument for Yandex management, listing risks and advantages. Once approved, there was history to preserve, packaging to build, a website and logo to arrange, documentation to expose and a release to explain. On June 15, 2016, ClickHouse opened under the Apache 2.0 license.
The invitation altered the project’s scale. Ten years later, ClickHouse counted more than 2,000 contributors. The database found its way into companies with problems its original team had never seen. In return, those users supplied bug reports, patches, demands and benchmarks. Milovidov, who says he collects benchmarks like other people collect stamps, had found a worldwide postal service.
His public work carries a mischievous strain. He has loaded aircraft transponder data into an interactive map, analyzed global weather from planes, visualized Foursquare places and built a playground where 110 database systems can be queried. In a talk on performance, his advice was admirably brief: “Do weird things, do real stuff.” The odd experiment is not a holiday from the product. It is a way to find where the product bends.
A company arrives after the obsession
In 2021, the core development team moved from Yandex into a new independent company. Milovidov co-founded ClickHouse, Inc. with Aaron Katz and Yury Izrailevsky, taking the CTO role. The combination was deliberate: the database creator, a commercial leader and an engineer experienced in large cloud systems. Their company would keep the open-source project and build managed services around it.
The founding happened during the pandemic. Katz and Milovidov worked together before meeting in person. When they finally did, in Amsterdam in October 2022, Katz expected a handshake. Milovidov seized him in a two-armed hug. This is a useful corrective to the chilly mythology of infrastructure software. A database may be columns, codecs and consensus. A company also has to be a place where its makers are glad to see one another.
ClickHouse Cloud launched in 2022. European offices opened in Amsterdam, where Milovidov’s public profiles now place him. The company grew, but his habits remained recognizable. He writes engineering posts, presents monthly releases, makes demos and contributes code. His LinkedIn feed in 2026 could still brighten at the prospect of faster CPUs, more cores, larger caches and greater memory bandwidth. Some executives acquire a taste for abstraction. Milovidov still enjoys the hardware aisle.
His recent interests show a builder willing to revise an old machine without pretending it should become a new one. ClickHouse is principally a C++ codebase, but its team began introducing Rust in carefully chosen places, starting with a hash function rather than a theatrical rewrite. The same pragmatism shaped his 2026 account of coding agents. He had been skeptical that an agent could manage the central C++ repository. After repeated success on bug investigations, small features and tightly specified tasks, agents joined the engineers’ daily work. New tools were admitted by evidence, not fashion.
That stance also explains why the public experiments matter. A map of aircraft or a race among 110 databases is entertaining enough to travel, yet each produces stubborn, unusual workloads. The demonstration becomes a question posed to the engine. Can you ingest this? Can you render that? What breaks when somebody tries the impolite query? Milovidov has spent a career turning such questions into test cases.
The 3 a.m. answer
The clearest explanation for his endurance came near the end of a long technical interview. Asked how he had kept the joy after 15 years, Milovidov described recognizing the solution to a problem and wanting to implement it immediately - at three in the morning, on a Saturday, if necessary. He simply goes and does it.
This appetite is not the same as believing every line deserves perfection. He cheerfully calls some temporary programs “trash scripts.” The distinction is part of the craft. Spend attention where attention changes the outcome. Measure before declaring victory. Keep production close. If a rival wins a benchmark, resist the poetry of excuses and inspect the profiler.
As ClickHouse enters its second decade in public, Milovidov’s stated ambition remains expansive: make it the first choice whenever people think about analytical databases, easier to use in the cloud and available to companies without specialist operators. Yet the story’s defining image is still small. A child types a BASIC instruction. The computer draws a circle. Wonder arrives because the response is immediate.
Everything since has been an argument for preserving that feeling at an absurdly larger scale.