EdgeDB: The Toronto Consultants Who Rewrote Python's Async Plumbing on the Way to Building a Database

The story

"So you've finally decided you're fed up with $database_name and simply have no choice but to build a new database," Yury Selivanov began, in a blog post looking back on the creation of EdgeDB. "All that's left is to quit your job and grind out a 1.0 in a few months. Right? Maybe! But—to put it mildly—that hasn't been our experience." The title of the post told the real story: "Building a production database in ten years or less."

**A boutique shop in Toronto**

The journey began in 2008, when Selivanov and Elvis Pranskevichus, his future EdgeDB co-founder, "started MagicStack, a boutique software development shop," he wrote. Over the years they worked with clients ranging "from early stage startups to Fortune 500 companies like General Electric and Microsoft." TechCrunch later noted that MagicStack was based in Toronto.

Like many consultants, they kept solving the same problems. "Early on, we realized we were solving the same set of problems over and over on every project. It was a drag, and it took time away from the fun and creative aspects of the work," Selivanov wrote. So they built their own toolkit, mostly in Python, including a component-based way to build declarative UIs, an RPC library and a bundler, and a data layer they called Caos. TechCrunch summarized the turning point: as they created bespoke tooling for clients, "the founders came to the realization that they wanted to lead a purely product-driven company as opposed to a consulting firm."

**Airbnb for boats**

Their first product attempt was not a database. "In Caos, we already had something exceptional on our hands, so naturally we decided to build…Airbnb for boats. Yes, really. It was called BoatBee," Selivanov wrote. Their tools let them build the product quickly. "Rather less quickly, we realized Airbnb for boats was a non-viable business. Live and learn."

The post-mortem changed their direction. Looking at the tools they had incubated for seven years, they saw that much of the ecosystem had caught up: React had made their UI kit obsolete, GraphQL overlapped with their RPC library, and Webpack beat their bundler. "But we saw no similar innovation in the database layer," Selivanov wrote. "In fact, it seemed to regress; we watched in dismay as schemaless NoSQL data stores exploded in popularity." Then came the insight: "Walking back from the office on a brisk Toronto evening, Elvis and I suddenly realized Caos could represent a new way forward. But first it had to stop being an ORM and become a database." The skeptical response they heard most often, "so it's just an ORM?", went up on their office wall for motivation.

**Fixing Python first**

Building a database server that could handle thousands of simultaneous connections required an asynchronous core, and Python's tools were not good enough. While designing a client library, the founders got stuck on how verbose transactions looked with the coroutine syntax of the time. "We realized this could be elegantly expressed if Python supported async with syntax natively. I started working on a PEP," Selivanov wrote. That work became part of Python itself. Vercel later credited him with contributing "async/await syntax to Python along with seven other Python enhancement proposals (PEPs)."

Speed was the next obstacle. In early 2016 the team combined Cython with libuv, the non-blocking I/O library from Node.js. "The result was uvloop: a drop-in replacement for Python's built-in asyncio event loop that was 2-4x faster," Selivanov wrote. "In surprisingly short order, uvloop grew to near ubiquity and now powers EdgeDB." When no existing asynchronous Postgres driver met their needs, "we built our own," he wrote, producing asyncpg. Vercel described uvloop as downloaded 3 million times per day and called asyncpg "deeply embedded in the Python ecosystem."

**Evenings at the whiteboard and a PyCon deadline**

With those foundations, the database itself took shape in 2017 and 2018. "While still consulting full-time, we would spend evenings at the whiteboard fleshing out designs," Selivanov recalled. To force a deadline, they signed up as a gold sponsor of PyCon 2018, five months away, and printed 3,500 booklets. "Naturally, we finished the preview implementation just under the wire, pulling an all-nighter at a random Airbnb in Cleveland," he wrote. The reaction convinced them: "People at PyCon loved EdgeDB."

EdgeDB was built on Postgres, which Selivanov called "the world's most advanced open-source SQL database," but he insisted it was more than an ORM, with "its own formally-defined query language, EdgeQL," its own type system, binary protocol and client libraries.

**From alpha to 1.0**

In late 2019, the founders "incorporated EdgeDB Inc. in the US and relocated our headquarters from Toronto to San Francisco." The company grew from three engineers to nine in 2020, released six alphas, and moved through betas and release candidates in 2021. On February 10, 2022, Selivanov and Pranskevichus announced EdgeDB 1.0, "the first open source, graph-relational database," with a query language "designed to surpass SQL in expressive power."

Later that year they announced a $15 million Series A led by Nava Ventures and Accel, with angels including Vercel's Guillermo Rauch, GitHub founder Tom Preston-Werner, Nat Friedman, Netlify's Mathias Biilmann Christensen, Firebase's James Tamplin and Sentry's David Cramer. Selivanov told TechCrunch the money would fund EdgeDB Cloud, because a database-as-a-service "requires significant investment upfront to build a reliable and scalable infrastructure." He argued that EdgeDB "reinvents pretty much every concept" about relational databases.

**Gel, and a bittersweet ending**

In February 2025 the company rebranded as Gel, with Pranskevichus writing that "PostgreSQL seems to be quietly eating the database world." Less than a year later, on December 2, 2025, Selivanov shared "bittersweet news": "Gel Data Inc. is shutting down, and our amazing team is joining Vercel to help build the best Python cloud in the world." Gel itself remained open source, with migration guides for users. Vercel said the team, "led by Python core developer Yury Selivanov and contributor Elvis Pranskevichus," would bring first-class Python support to its platform.

EdgeDB is an honest story about how long ambitious infrastructure takes and how outcomes can diverge from intentions. Two consultants spent years discovering what they really wanted to build, detoured through a failed marketplace, and then had to improve their programming language before they could build their database. Those detours produced some of Python's most important async infrastructure, used by millions of developers who may never touch EdgeDB. The company did not become a standalone database giant, but its founders' open source contributions outlived the business and became the reason a major platform wanted the team.