It's 2:14 a.m. Someone in Berlin can't check out.
A Shopify-sized cart is hanging. The mobile app is freezing on a single button. Somewhere in San Francisco, a phone buzzes - and an engineer who hasn't opened her laptop yet already knows three things: which release shipped the bug, which line of TypeScript threw it, and what the user was doing in the seventeen seconds before the screen froze.
She knows all of that because Sentry told her. Not the logs. Not Slack. Not a customer-success ticket that will land in her inbox on Tuesday. Sentry.
That is the company in 2026: a piece of infrastructure quietly threaded through more than 100,000 organizations and four million developers' workflows. It is not glamorous. It does not trend. It simply sits there, watching production, and when production misbehaves it produces something rare in software - a sentence that begins with “here is exactly what happened.”
Software has always shipped broken. We just stopped pretending it didn't.
Every developer alive has typed the phrase “works on my machine.” Most of them have meant it. The trouble has always been the gap between a developer's machine and a production server in some other timezone holding the credit card of a customer who, at this moment, is about to leave a one-star review.
Before Sentry, the standard response to that gap was a heroic act of detective work: SSH into a box, tail a log, grep for something that resembled an error, and hope the timestamps lined up. Junior engineers learned the dance. Senior engineers wrote scripts to dodge it. Nobody enjoyed it.
The deeper problem - the one Sentry's founders saw earlier than most - is that errors in production aren't edge cases. They are the actual job. A modern app throws hundreds of exceptions a minute. Most are harmless. A few are catastrophic. Sorting them by hand is a fool's errand.
Forrester estimates a single hour of customer-facing downtime can cost an enterprise more than $300,000. The smaller, slower bleed - intermittent crashes that drop conversion by 1 or 2 percent - is, in aggregate, larger and almost impossible to find without instrumentation.
A Django side project, a Heroku add-on, and a refusal to ask permission.
David Cramer is not the person you'd cast for “CTO of unicorn.” He dropped out of high school in 9th grade. He managed a Burger King. He taught himself Python because asking the systems administrator for log access took too long, and David - by his own description - is the kind of person for whom “too long” is an existential problem.
The first version of Sentry was, by Cramer's own telling, the laziest possible solution. If he couldn't get to the logs, he would make the logs come to him. He wrote it for himself, open-sourced it, and the Python community quietly adopted it. By 2012, he had recruited Chris Jennings - an engineer with stops at GitHub and Disqus - and turned the side project into a company. They launched it as a Heroku add-on over Christmas vacation. They charged for it. People paid.
Cramer and Jennings bet that developers would buy tools with their own credit cards if those tools were useful enough - a heresy in an industry that, at the time, still believed software was sold to CIOs in steakhouses.
That heresy aged well. Sentry was profitable before it took venture money. Accel led a $1.5M seed in 2015 mostly so the founders would stop saying no. The company has since raised $219 million across six rounds, capped by a $90M Series E in May 2022 at a $3 billion valuation - and along the way installed Milin Desai, the former GM of VMware's cloud business, as CEO so Cramer could go back to doing what he liked best: arguing with engineers about product.
What Sentry actually does, in five sentences and a screenshot you don't have.
Drop a small SDK into your app. Sentry catches every unhandled error - and, increasingly, every slow transaction, every crashed mobile session, every flaky cron job. It groups them by fingerprint, so an exception thrown 40,000 times in one hour appears as one issue, not 40,000. It links the issue to the exact commit that introduced it. It can play you a video of the user's session leading up to the bug. And, since 2024, it can write a suggested patch.
The platform, unpacked
The original product. Over 100 languages and frameworks. Stack traces, breadcrumbs, release health, and a release-comparison view that's arguably its quietest superpower.
Distributed tracing across services. Sentry surfaces slow database queries, hanging API calls, and the thirty-millisecond regressions that, multiplied across a million sessions, turn a launch into an emergency.
A pixel-accurate recording of what the user actually did. Engineers love it because it ends an argument that has gone on for two decades: “I can't reproduce it.” You can now. Hit play.
Continuous code-level profiling shows you the function that's eating 14% of CPU. Cron monitoring catches missed jobs. Codecov - acquired in 2022 - folds test coverage into the same workflow.
The Sentry decade, in dotted lines.
-
2008David Cramer open-sources the first version of Sentry as part of his work in the Django community. No company. No plan. Just an itch.
-
2012Cramer and Chris Jennings turn the project into a business. Sentry ships as a Heroku add-on. First paying customers within weeks.
-
2015 - 2016Seed round led by Accel ($1.5M). Series A follows in 2016 ($9M, NEA leads).
-
2019Series C of $40M. Sentry expands beyond errors into performance monitoring.
-
2021$60M Series D at a $1B valuation. Officially a unicorn. Milin Desai joins as CEO.
-
2022$90M Series E at a $3B valuation. Sentry acquires Codecov. Session Replay enters beta.
-
2023Sentry introduces the Functional Source License (FSL) - a balance between open source and commercial protection that quietly nudges a debate the industry has been ducking.
-
2024 - 2025AI-assisted issue triage and suggested fixes ship. Continuous profiling goes GA. Session Replay expands to native mobile.
The numbers, and the names that signed off on them.
If a product is good, the strongest evidence is who chooses to depend on it. Sentry's customer list reads like the org chart of the internet: Disney, Microsoft, Atlassian, GitHub, Cloudflare, Slack, Peloton, Riot Games, VMware. None of those companies are short on engineers. They use Sentry anyway, because building the equivalent internally is a project no one wants to staff.
Funding, round by round
USD millions raised, 2015 - 2022
Partnerships worth naming
The product lives where developers do. GitHub integrates Sentry issues directly into PRs and commit links. Vercel offers a one-click install for Next.js. AWS distributes it through Marketplace. Heroku - the original distribution channel - still ships it as an add-on, more than a decade later, which is a kind of compliment.
Open enough to be loved. Commercial enough to be sustainable.
Sentry has built something rarer than a product: a stance. The company spent its first decade quietly funding the open-source projects it depended on - publishing what it calls “open source dividends,” a share of revenue paid back to maintainers. Then, in 2023, it did something braver and noisier. It moved core repos to a new license it co-authored, the Functional Source License (FSL), which keeps the code freely readable, freely forkable, and freely usable by everyone except the handful of companies who might want to resell Sentry as Sentry.
The internet, predictably, argued. Some called it a betrayal of open source. Others called it the only honest answer to a decade of cloud providers extracting commercial value from community-maintained software. The license, like most things Sentry does, was less interested in being polite than in being clear.
Customer obsession. Craftsmanship over completeness. A bias toward shipping. A weakness for arguing in public. A documented refusal to chase metrics that don't move the product.
An AI wave is coming for software. Someone still has to watch the wave.
Most of the AI conversation is upstream - models, agents, copilots, the things that write the code. Sentry's bet is downstream. The code being written today by a Claude or a Cursor or a Devin is being written faster than humans can fully read it. It will ship to production with bugs no one understood while typing. The discipline of watching that code behave in the real world - and explaining what it does when it misbehaves - is about to matter more, not less.
Sentry's AI features are unflashy on purpose. Triage suggestions. Suggested fixes. A “here's what probably broke” sentence above the stack trace. They are not trying to replace the engineer. They are trying to make the engineer faster at 2:14 a.m., which is when the engineer is least replaceable.
Closing the loop
Back to Berlin. The engineer in San Francisco doesn't write a long postmortem. She doesn't ping the customer-success team. She opens Sentry, watches the replay, sees that a feature flag flipped a button into a state nobody tested, and rolls back the release. The cart unhangs. The user in Berlin - the one who was three seconds away from a one-star review - completes checkout and never notices anything happened.
That is the trick Sentry has been pulling off, quietly, for more than a decade. It is not the loudest company in San Francisco. It is not even the loudest company on its block. It is simply the one a great many developers reach for first, when the night gets long.