DuckDB started at CWI in 2018
DuckDB lets an analyst query sales data on a laptop without setting up a separate database server. It runs inside the program using it.
In 2018, Mark Raasveldt and Hannes Mühleisen were researchers in the Database Architectures group at CWI, the Dutch national research institute for mathematics and computer science. They had already built MonetDBLite, an embedded version of the group’s MonetDB database. In a 2019 paper, they wrote that it had thousands of downloads a month, with users ranging from the Dutch central bank to the New Zealand police, but that its success had also uncovered problems that were very hard to fix in a system not built for the job.
In a 2023 interview republished by CWI, Mühleisen said that of four possible kinds of database, only three existed. Databases were built either for analysis or for transactions, and ran either as a separate server or inside a program; none combined analysis with running inside the program. “So we set out to develop a prototype in 2018.”
The repository was created on 26 June 2018 under the CWI group’s GitHub account. Raasveldt made the first commit on 13 July: a working SQL parser and a draft interface. On 1 August, a commit reported that TPC-H Q1, a standard benchmark query, was “mostly working”. The first preview release followed on 27 June 2019, days before SIGMOD, a major database research conference held that year in Amsterdam, where their demonstration paper appeared.
They released DuckDB as open source. “We find it unethical to make proprietary software based on taxpayer funded research,” Mühleisen said.
Seven Hacker News submissions drew 6 points or fewer
“Most new Open Source projects never gain any traction, so that was also our expectation in the beginning,” Mühleisen said in 2024. Before DuckDB, he and Raasveldt had built a “hacky prototype” that got some people excited, which gave them the confidence to start from scratch.
The repository has 409 surviving stars (stars not since removed) dated 2019. July accounts for 175 of them, in the weeks of v0.1.0’s release and the SIGMOD demonstration; from August 2019 to April 2020, no month had more than 50. Releases continued: seven preview releases, from v0.1.0 in June 2019 to v0.1.7 in May 2020, monthly from January 2020.
Hacker News users submitted DuckDB seven times in those 11 months, and no submission drew more than 6 points. None came from the creators’ Hacker News accounts. The first, on 30 June 2019, used the README’s heading, “DuckDB, the SQLite for Analytics”, and got 3. Two were blog posts by early users trying the database. The seventh went up on 23 May 2020, less than an hour before the post that took off, and also got 3.
Stars doubled after a Hacker News post in May 2020
Up to 22 May 2020, 543 surviving stars are dated to the repository. At 3:47 pm US Pacific time on 23 May 2020 (22:47 UTC), a Hacker News post titled “DuckDB: SQLite for Analytics” linked straight to the GitHub repository. It reached 290 points and 67 comments. The title was almost the same as the first submission’s, which echoed the heading of the repository’s README, in place since 2018. The founders’ 2019 paper made the same comparison, opening with SQLite’s popularity and noting that no such system yet existed for analysis.
In the seven days from the post, the repository gained 599 surviving stars, against 5 in the week before. By the end of May, the total had more than doubled, to 1,154. The timing ties the jump to the post, though star data cannot show where each person found the repository. Raasveldt and Mühleisen replied to questions from the next morning, Amsterdam time; 6 of the thread’s 67 comments are theirs.
| Stars dated to each day, still present at collection | 16: 1 | 17: 2 | 18: 2 | 19: 0 | 20: 0 | 21: 0 | 22: 0 | 23: 255 | 24: 216 | 25: 67 | 26: 23 | 27: 6 | 28: 16 | 29: 16 | 30: 4 | 31: 8 |
|---|
GitHub star history for duckdb/duckdb, 16–31 May 2020, collected on 29 October 2026. Removed stars are not counted.
The pattern repeated four months later. At 5:02 pm Pacific time on 19 September 2020 (00:02 UTC on 20 September), a second post linking to DuckDB’s website reached 801 points. The repository gained 609 surviving stars in the following seven days, against 6 in the week before. Raasveldt and Mühleisen wrote 18 of that thread’s 160 comments.
Python downloads grew later and kept growing
Stars record attention. Package downloads are closer to use, though they also count automated installs, such as test servers installing a package on every run.
DuckDB’s early Python downloads were mostly from mirrors: tools that copy packages from PyPI, the Python package index, often every package, whether or not anyone uses them. Mirror tools account for 81% of DuckDB’s recorded PyPI downloads in 2019 and 79% in 2020, falling below 1% by 2025. The figures here exclude them, along with the small share that records no installer.
Python downloads did not rise after the May 2020 Hacker News post: 152 in the seven days from the post, against 353 the week before. Downloads rose from mid-June: 1,374 in May, 6,052 in June and 30,150 in July, then 23,000 to 28,000 a month through October. Most of June’s rise was browser downloads; installs through pip jumped in July. The data does not show who was downloading DuckDB or why.
| Year | Stars | PyPI | npm |
|---|---|---|---|
| 2019 | 409 | 2.1k | – |
| 2020 | 1,868 | 156k | 3.6k |
| 2021 | 1,609 | 679k | 84k |
| 2022 | 3,591 | 4.1M | 337k |
| 2023 | 5,715 | 16.1M | 1.6M |
| 2024 | 11,627 | 59.4M | 3.7M |
| 2025 | 9,855 | 219M | 13.4M |
Stars are surviving stars dated to each year; other columns are downloads. 2026 is omitted as a partial year. PyPI excludes mirror tools and starts on 8 May 2019. npm, the JavaScript package registry, combines the duckdb, @duckdb/duckdb-wasm and @duckdb/node-api packages; DuckDB’s npm package first appears in November 2020.
Python carried most of the downloads. In 2021, the year Labs was founded, PyPI recorded about 679,000 downloads, against about 84,000 from npm and 57,000 from CRAN, R’s package archive, which first lists DuckDB downloads in September 2020 and counts only the Posit mirror. Stars and downloads also moved apart: fewer surviving stars are dated 2021 than 2020, and fewer 2025 than 2024, while downloads grew in both years.
Labs chose contracts over venture capital
By July 2021, CWI reported, DuckDB was downloaded about 100,000 times a week. CWI did not say how it counted. In the same weeks, PyPI, npm and CRAN together recorded about 14,000 to 16,000 downloads a week excluding mirrors, or roughly 50,000 to 80,000 including them. That month, CWI announced that Mühleisen and Raasveldt had founded DuckDB Labs as a spin-off to develop and support the database.
Venture capital firms were calling, the founders later wrote, but they kept the company bootstrapped. Mühleisen explained the choice in 2024: because DuckDB started in academia, it had gained traction before the company existed, and customers were already lined up to pay for consulting and support. Amsterdam’s lower operating costs helped too.
As Mühleisen described it, customers paid Labs to build features sooner or to add ones that were not on the roadmap. Every feature went into the single open-source version.
Tigani started MotherDuck in 2022
According to TechCrunch, Jordan Tigani, a founding engineer on Google BigQuery, was chief product officer at SingleStore when he noticed that most database workloads were small, under 10GB. He contacted Mühleisen to discuss a paid service built on DuckDB.
When Tigani first met Mühleisen and Raasveldt in April 2022, they told him they had already been turning down acquisition offers, he later wrote. With Mühleisen’s blessing, he began building MotherDuck, a cloud service around DuckDB. It was designed to let users query local and cloud data together, with the service deciding where to perform the work.
How Andreessen Horowitz explained the Series A
On 15 November 2022, MotherDuck announced $47.5 million in funding. The total covered seed and Series A rounds, reported by TechCrunch as led by Redpoint and Andreessen Horowitz respectively. Tigani told TechCrunch that MotherDuck had not planned to raise a Series A so soon after its seed, and did so at the urging of LPs, the investors behind venture funds, and for the chance to work with Andreessen Horowitz.
Andreessen Horowitz partners Martin Casado and Jennifer Li explained the investment starting with the market. Most analytics tools, they argued, were built to process petabytes across many machines. For smaller jobs, which could still reach many terabytes, those tools were complicated and slow. They called that mismatch one reason for DuckDB’s “absolutely explosive growth” over the previous year.
They wrote that hundreds of thousands of data scientists and analysts downloaded and used DuckDB, but that being a downloaded program limited its potential to be driven from cloud programs and shared between users. In their view, MotherDuck’s hosted service addressed that limit.
They also pointed to Tigani’s decade building BigQuery, to engineers from Snowflake, Databricks, AWS, Meta, Elastic and Firebolt, and to the partnership with DuckDB Labs, the database’s developers.
The post cited no revenue or customers, and Tigani declined to give TechCrunch either figure, saying it was too early. Andreessen Horowitz’s stated case rested on DuckDB’s usage and the team’s plan for a cloud service.
Contracts and shares connected the businesses
On the same day, DuckDB Labs announced a long-term partnership with MotherDuck. MotherDuck joined partners such as Voltron Data and RelationalAI that paid Labs for ongoing DuckDB development. Labs said it was owned only by its founding team and CWI.
Labs also acquired shares in MotherDuck. Its stated reason was that consulting alone “does not scale well”; the shares tied Labs’ success to its partner’s. MotherDuck began supporting the nonprofit DuckDB Foundation, which holds the project’s intellectual property and keeps it under the MIT open-source license.
On Hacker News, an essay outscored MotherDuck’s launch
On 7 February 2023, before its service opened, MotherDuck published Tigani’s essay “Big data is dead”. It argued that most companies’ data was small enough for a single machine. On Hacker News, it drew 879 points and 433 comments, the most of any MotherDuck post there.
MotherDuck’s launch on 22 June 2023 drew 17 points as a Show HN post and 24 in a separate submission of its blog post. The Hacker News posts about its two funding rounds each drew fewer than 50.
The Series B backed a service that was still free
On 20 September 2023, MotherDuck announced a $52.5 million Series B led by Felicis, bringing its total funding to $100 million. It also removed the service’s waitlist.
MotherDuck reported that 2,000 users had queried the service since its launch three months earlier. Those were not reported as paying customers: the service remained free, with billing planned for the following year. Felicis’s public explanation emphasized the team’s database experience and local-plus-cloud architecture.
Neither announcement month stood out in surviving GitHub stars. November 2022 had 423, below September’s 448 and October’s 494; September 2023 had 476, between August’s 390 and October’s 556. DuckDB releases also fell nearby, v0.6.0 the day before the first announcement and v0.9.0 six days after the second, so star counts in those months cannot be tied to the funding news.
DuckDB reached 1.0, and MotherDuck reached general availability
Until 2024, database files written by one DuckDB version could not always be opened by the next and had to be upgraded manually. February 2024’s v0.10.0 introduced support for reading files from v0.9. With version 1.0 on 3 June 2024, the maintainers promised that files created with 1.0 would remain readable by future versions. June 2024 had 2,671 surviving stars, the most of any month, though the rise began before release day.
The 1.0 announcement said consulting and support contracts paid for almost 20 people at Labs.
Eight days later, MotherDuck announced general availability, listing paid plans alongside a free one. It said many customers were already running the service in production.
Amazon acquired Labs in 2026
On 26 August 2026, Amazon announced an agreement to acquire DuckLabs, as DuckDB Labs had been renamed that May. The founders described the deal as joining AWS. Amazon said it was not acquiring the DuckDB project itself, which stays under the MIT license with the DuckDB Foundation. The deal closed on 31 August.
By then, the founders wrote, the company had more than 30 employees in Amsterdam, and DuckDB was downloaded more than a million times a day. They worried that DuckDB’s growth would outpace their small company, and that building a much larger sales, support and operations organization would pull them away from technical work.
Tigani wrote the same day that MotherDuck would now offer enterprise DuckDB support, which it had avoided while Labs depended on that business. He also expected Amazon to launch its own DuckDB-based service eventually.
What a small team can take from DuckDB and MotherDuck
We read DuckDB’s early record as a case study for teams launching open-source tools. These are our inferences from the evidence above, not the founders’ claims.
Early Hacker News scores were a poor verdict on the project. The same “SQLite for analytics” wording drew 3 points in June and December 2019 and 290 in May 2020, and the user who posted the 290-point submission drew 3 points with another DuckDB link that August. The record does not show why one submission took off and the others did not.
MonetDBLite showed demand before DuckDB’s first commit: thousands of monthly downloads and institutional users. Its limits became DuckDB’s requirements.
“SQLite for analytics” described the database by naming one readers already knew, and it headed the README from 2018.
The week after the May 2020 Hacker News post saw a jump in stars but no rise in Python downloads, and downloads kept growing in 2021 and 2025, years with fewer new stars than the year before. Stars alone would have been a poor guide to use.
Labs’ first revenue came from companies already relying on DuckDB, which paid to have features built sooner; every feature went into the open-source version, so paid work did not split the product.
For MotherDuck, an argument drew far more Hacker News votes than its announcements: 879 points for Tigani’s essay against 24 or fewer for its launch.
Sources and method
This article covers 2018–2026. GitHub’s repository star history API, collected on 29 October 2026, counts only stars still present (“surviving stars”), not historical displayed totals or users. Daily, monthly and yearly groupings use the returned week timestamps and day offsets. GitHub’s day boundaries appear to follow US Pacific time: 119 stars are dated 19 September 2020, before the second post went up at 00:02 UTC on 20 September. Commit dates come from the duckdb/duckdb Git history; the repository was first hosted under CWI’s cwida account, which now redirects there.
Hacker News points and comments come from the Algolia Hacker News API in October 2026. The founders’ comments are those from mytherin (profile: “Lead developer of the DuckDB database system”) and hfmuehleisen (profile: “DuckDB”); in the threads, mytherin refers to “Hannes and me” and hfmuehleisen writes as one of DuckDB’s authors. Comment totals are Hacker News’s counts; 157 of the September thread’s 160 comments are retrievable, and founder counts come from those. PyPI downloads come from ClickHouse’s public ClickPy dataset, built from PyPI’s download logs. We excluded downloads by mirror tools (bandersnatch, devpi, Artifactory, Nexus and z3c.pypimirror) and those with no recorded installer or the generic OS label; our September 2026 total matched pypistats.org’s count without mirrors to within 0.1%. npm figures come from the npm downloads API, and CRAN figures from the cranlogs API. Download counts include automated installs and do not measure users.
Funding dates are announcement dates, not necessarily closing dates. Usage, download and staffing figures attributed to CWI, DuckDB Labs, MotherDuck or the founders are as they reported them. We found no MotherDuck funding announcement after September 2023; third-party databases list a 2025 round we could not confirm.