GitHub, explained: where the world's software actually lives

The short answer

GitHub is a platform for hosting and collaborating on code - the largest one in the world by an enormous margin. It combines Git version control, code review, issue tracking, CI/CD automation, and an AI pair-programmer into one place, and it is where the overwhelming majority of open-source software lives.

The scale is hard to overstate: over 180 million developers, roughly 630 million repositories, and nearly a billion commits pushed in 2025 alone - about one new developer joining every second. Whatever your site or app is built from, parts of it almost certainly passed through GitHub.

Git is not GitHub

The most common confusion is between the tool and the platform. Git is the version control system itself - created by Linus Torvalds in 2005 to manage the Linux kernel - that tracks every change to a codebase so nothing is ever lost and every version can be recovered. It is free, open source, and works entirely offline.

GitHub is a hosted service built around Git that adds the collaboration layer: the place where repositories live online, where changes get proposed and reviewed, and where the work of building software gets organized. Think of Git as the engine and GitHub as the city built around it.

Git itself: the tool underneath

Git has its own origin story worth knowing. In 2005, Linus Torvalds needed a version control system for the Linux kernel after the kernel project lost its free license to BitKeeper, the commercial tool it had been using. Unsatisfied with every existing option - they were too slow and too centralized for thousands of kernel contributors - he wrote his own in a matter of weeks. The name is self-deprecating British slang for an unpleasant person; Torvalds has joked that he names projects after himself: "first Linux, now Git."

The design choices that made it work for the kernel are why it won everywhere. Git is distributed: every clone contains the entire history, so there is no single point of failure and work continues offline. Commits are snapshots of the whole tree, not diffs, identified by a content hash - which makes history tamper-evident. Branches are just movable pointers to commits, which is why branching and merging - painful in older systems - became cheap enough to do constantly. And the staging area lets you compose exactly what goes into each commit.

Those primitives are the foundation under everything else in this article: the pull request is a request to merge one branch into another, a review is a conversation about a diff between snapshots, and CI is automation that reacts to commits. GitHub did not invent version control; it made Git's model legible to everyone.

The founding story

GitHub was founded in February 2008 in San Francisco by Tom Preston-Werner, Chris Wanstrath, PJ Hyett, and later Scott Chacon - a group of developers who saw that Git was powerful but had a miserable user experience and no good way to collaborate on it. Their insight was social: put a profile and a feed on code, and contributing to a project becomes as easy as a conversation.

Famously, the company was bootstrapped - it took no outside money for its first four years, then in 2012 accepted a single $100 million investment from Andreessen Horowitz, at the time the largest round the firm had ever done. By then it had already become the de facto home of open source.

Octocat, stars, and the social layer

Part of GitHub's staying power is that it turned a dry tool into a culture. The mascot - the Octocat, a cat-octopus hybrid - came from stock art by illustrator Simon Oxley that the founders licensed early on, and grew into one of tech's most recognizable logos with its own merch culture. Stars, followers, and profile pages gave developers a public identity attached to their work; the green contribution graph became a de facto resume.

That social layer mattered strategically. A GitHub profile became something closer to a portfolio than a login - employers read it, contributors were discovered through it, and the README file evolved into the homepage of a project. The platform layered funding on top later with GitHub Sponsors in 2019, letting people pay maintainers directly - an attempt to fix open source's oldest problem, that the whole ecosystem ran on volunteers.

The pull request changed open source

GitHub's single most important invention was not a feature but a flow: fork, change, pull request, review, merge. Before it, contributing to an open-source project meant emailing patches to a maintainer. GitHub turned it into a visible, linkable conversation - anyone could propose a change, and the whole discussion lived next to the code.

That flow is why open source exploded the way it did. It lowered the social friction of contributing at exactly the moment the industry was moving to open source as its default supply chain - and it quietly set the pattern for how nearly all software, open or closed, gets reviewed today.

The rough patches

The history has real scars worth naming. In 2014, co-founder and then-former CEO Tom Preston-Werner resigned after an internal investigation into harassment allegations by engineer Julie Ann Horvath; the investigation found errors of judgment rather than the alleged harassment itself, but the episode forced an early reckoning with the company's culture. A year earlier GitHub had weathered state-level censorship and blocks in several countries, and in 2015 it absorbed a massive DDoS attack widely attributed to China's Great Cannon.

The attacks kept scaling with the platform. In February 2018 GitHub survived a 1.35 terabits-per-second DDoS - the largest publicly recorded attack at the time - without going down. A different kind of turbulence came in 2019, when hundreds of employees signed an open letter protesting the company's contract renewal with US Immigration and Customs Enforcement, a fight that foreshadowed the industry-wide debate over who tech companies sell to.

The Microsoft era

In June 2018, Microsoft acquired GitHub for $7.5 billion in stock - a move that would have seemed absurd a decade earlier, when the companies were seen as opposites, and that turned out to be one of the savvier acquisitions of the cloud era. Microsoft left it largely autonomous under CEO Nat Friedman, then Thomas Dohmke.

Under Microsoft the platform deepened rather than changed character: GitHub Actions became a major CI/CD system, the company bought npm (the JavaScript package registry), Dependabot, and Semmle - whose CodeQL engine became the analysis behind code scanning - shipped Codespaces for cloud development environments, and in 2025 folded GitHub into Microsoft's CoreAI organization as the platform's AI layer became the center of gravity. The open-source community's initial skepticism mostly gave way to the reality that the platform kept improving.

The products that came and went

Not everything GitHub touched survived, and the failures are instructive. Atom, the hackable text editor GitHub launched in 2014, was a beloved experiment that never beat Microsoft's VS Code - and was quietly sunset at the end of 2022 after the acquisition made the overlap awkward. Yet Atom's real legacy is enormous: it shipped with a web-technology shell called Atom Shell, which became Electron - the framework that now runs VS Code, Slack, Discord, Figma's desktop app, and much of the software on your computer. GitHub's dead editor turned out to be the substrate of an entire generation of desktop apps.

Others endured in quieter forms: Gists for sharing snippets, GitHub Pages for hosting sites straight from a repository, a desktop app, and the `gh` command-line tool that brought the whole platform into the terminal - a fitting loop for a tool that began as a friendlier face on a command-line system.

What lives on the platform

The feature surface is broad because the job is broad - hosting a codebase is only the start:

• Repositories - hosted Git with the full history of every change, branch, and tag
• Pull requests - the review surface where changes get discussed, approved, and merged
• Issues and Projects - the tracking and planning layer where work is organized
• Actions - CI/CD automation that builds, tests, and deploys code on every push
• Codespaces - full cloud development environments that open in a browser
• Pages, Packages, Releases - hosting for sites, dependencies, and shipped versions
• Discussions, wikis, Gists - the community and documentation layer around the code

Actions: the automation engine

GitHub Actions, launched in 2018 and generally available in 2019, turned the repository into an automation platform. A YAML file in the repo defines workflows that run on events - push, pull request, schedule - building, testing, and deploying the code automatically. A marketplace of community-written actions covers almost any task.

The scale is staggering: over 11 billion Actions minutes were spent just running tests in 2025, and the platform handles tens of millions of workflow runs daily. For most modern projects, the CI pipeline is the repository - which is also why an Action that lints, typechecks, and builds on every pull request has become the standard guardrail of a professional codebase.

Copilot: the AI layer

GitHub Copilot launched as a technical preview in 2021 and went generally available in 2022 - the first mainstream AI pair-programmer, built on OpenAI's Codex model. It started as autocomplete for code and grew into a multi-model assistant: inline suggestions, chat, a CLI, code review, and an agent mode that can take an issue and open a pull request on its own.

The numbers explain the impact: more than 20 million developers have used it, over three billion code suggestions have been accepted, and in GitHub's own 2025 data, 80 percent of new developers on the platform use Copilot within their first week. A free tier launched in late 2024 brought in the next wave - the same wave that powers the vibe coding movement, since Copilot's agent mode is one of its main workbenches.

A quieter but significant shift happened at GitHub Universe 2024: Copilot went multi-model, letting developers pick between OpenAI, Anthropic Claude, and Google Gemini models instead of being locked to one provider - an acknowledgment that the model layer commoditized and the platform layer is where GitHub's value sits. Copilot now ships in tiers from Free through Pro, Pro+, Business, and Enterprise, spanning inline completions, chat, code review, the CLI, and a coding agent that takes a GitHub issue and returns a pull request.

Spark and Models: the AI surface beyond Copilot

GitHub's AI reach extends past the editor. GitHub Models is a built-in playground for trying and comparing models - OpenAI, Meta, DeepSeek, xAI and others - directly against your code, without leaving the platform or managing keys.

Spark is the bigger swing: announced at Universe 2024 and in public preview for Copilot Pro+ users since July 2025, it builds full-stack apps entirely from natural language - powered by Claude Sonnet - with data, LLM inference, auth, hosting, and deployment included out of the box. Each iteration saves versions, can open into a Codespace or a repository, and hands off to the Copilot coding agent. It is, functionally, GitHub's own answer to the vibe coding builders - and it makes the platform's ambition plain: the repository is where the idea lands, whatever produced it.

The security layer

Because so much of the world's code lives there, GitHub also became a security surface. Dependabot watches dependencies for known vulnerabilities and opens fix pull requests automatically. Secret scanning catches leaked API keys and credentials before they spread. CodeQL performs deep semantic analysis for real vulnerabilities. Advisories and the dependency graph let a project see its whole software supply chain.

These are the same classes of risk the vibe coding article covers from the other direction - most of the ways AI-generated code fails are things this layer exists to catch, which is why it matters more, not less, as more code gets generated faster.

The surrounding ecosystem

Around the product is a deliberately cultivated ecosystem. GitHub Universe, the annual developer conference, is where the platform's direction gets announced each fall - the multi-model Copilot and Spark reveals both happened there. GitHub Skills and Classroom teach Git and development workflows, and the Student Developer Pack bundles free tools and credits that have onboarded a generation of students to the platform early.

Octoverse, the annual data report, doubles as the industry's scoreboard - the source of the numbers in this article. And GitHub Sponsors keeps trying to solve the maintainer-funding problem. Taken together it is a flywheel: teach people early, give them a public profile, give projects a way to get funded, and the platform keeps its position at the center of it all.

GitHub by the numbers

GitHub's annual Octoverse report is the clearest measure of where software is going, and the 2025 edition captured the AI shift in hard numbers: 180 million developers, 630 million repositories with 121 million created in the year alone, nearly a billion commits pushed, and over half a billion pull requests merged. A new developer joins more than once a second.

The telling details are in the trends: TypeScript overtook both Python and JavaScript as the most-used language in August 2025 - a shift attributed directly to AI-assisted coding favoring typed languages - and more than 1.1 million public repositories now use an LLM SDK. The platform is no longer just where code is stored; it is where the industry watches itself change.

Open source's home and its critics

GitHub is effectively the public square of open source, which brings real responsibilities and real criticism. Centralization is the biggest worry: when one company hosts nearly all of open source, an outage or a policy decision affects everyone, and alternatives like GitLab and Codeberg exist partly as insurance. The Arctic Code Vault project - GitHub archiving public repositories on film in a Norwegian vault - is an acknowledgment of how much civilization's software heritage now sits on the platform.

The other live controversy is Copilot itself: it was trained on public repositories, including code under licenses that require attribution, and a class-action suit over that was filed in 2022. Most claims were dismissed by 2024, but the underlying question - what obligations generated code carries from its training data - remains legally unresolved, which is exactly the gray zone the vibe coding piece covers.

Pricing and the competition

The core platform is free - unlimited public and private repositories, with Actions minutes and other limits on the free tier. Paid plans add features for teams (around $4 per user per month) and enterprises (around $21 per user per month for advanced security, compliance, and support). Copilot is priced separately across Free, Pro, Pro+, Business, and Enterprise tiers - Pro+ and Enterprise unlock the agent and Spark features. For the largest organizations, GitHub Enterprise Server offers a self-hosted version, and Advanced Security adds code scanning and secret protection to private repos.

The competition is real but smaller. GitLab is the most complete alternative, stronger for teams that want an integrated DevOps platform or self-hosting. Bitbucket serves the Atlassian ecosystem. Codeberg and others offer community-governed hosting. But GitHub's gravity - the network of developers, the integrations, the visibility of open source - is a moat no competitor has closed.

GitHub in the agent era

The platform's newest role is as the workplace for AI agents, not just humans. Copilot's coding agent takes a GitHub issue, works in a cloud environment, and opens a pull request for review - turning the repository into the queue where agents pick up work. GitHub also ships an official MCP server, so external AI tools can read issues, PRs, and workflows directly. The repository has become the point of coordination between human and machine work on the same codebase.

How this site uses it

Every project we ship lives on GitHub. The workflow is the standard one this article describes: feature branches, pull requests that build Vercel preview deployments, a review, a merge, a production deploy - all traceable to a commit. The client's codebase is a standard Git repository they can clone, fork, or hand to another developer at any time; nothing about it is proprietary to us.

That ownership point matters more than any feature: the code on GitHub is the deliverable, not a lease on it. If you want to see how a real project flows through this pipeline, the website development page covers it - or book a call.