GitCafe, a San Francisco startup that recently emerged from Y Combinator, opened its public beta this week with a pitch that sounds almost quaint in its ambition: build a better alternative to GitHub. The company claims hundreds of users already on the platform, though it's staffed by just two people.
What makes the bet less quaint is the specific problem GitCafe is trying to solve. Founder Veronica C. isn't chasing the mass market of developers happily using Microsoft-owned GitHub for pull requests and code reviews. She's designing infrastructure for a future where software teams ship hundreds or thousands of changes daily, many of them authored not by humans but by AI coding agents.
"We're building for the next hundred billion commits and CI jobs," Veronica C. wrote on the company's Y Combinator directory page. Whether that future materializes as quickly as she expects remains an open question, but recent research has identified AI-authored pull requests scattered across tens of thousands of commits on GitHub from tools like Copilot, Devin, Cursor, and Claude Code. The trend, at minimum, is real.
Rebuilding Git From Scratch
GitCafe is what developers call a "git forge," the technical term for platforms that host code, manage pull request reviews, and trigger continuous integration pipelines. But rather than wrapping existing Git tools, Veronica C. and her co-founder chose to write their own Git implementation from the ground up. The company lists four components: libgitz (the Git engine), kiln runtime, cafe:git native binding, and the forge itself.
It's an audacious technical choice, one that signals either ambition or hubris depending on how charitable you're feeling. Veronica C. was previously a founding engineer at Corgi, another Y Combinator company, and she doesn't oversell what GitCafe has built so far. "The honest version is that GitCafe is behind GitHub today," she wrote in the launch post. "We expect to earn the broader product surface, not claim it on day one."
The platform includes features designed for high-velocity development: push-time secret scanning, live merge-readiness checks, and passwordless authentication via passkeys. It supports stacked diffs through integrations with Jujutsu and Sapling, two version control systems that let developers layer multiple changes atop one another. GitCafe connects to Buildkite and Harmont CI, an agent-focused continuous integration platform, with beta support for GitHub Actions. GitLab CI support is planned.
There's also a command-line interface that handles authentication, pull request creation, and what the company calls "agent skills," installable capabilities designed explicitly for AI coding agents. One command: cafe skills install.
The Crowded Field
GitCafe is entering a market with entrenched incumbents and well-funded challengers. GitHub dominates code hosting with features like Actions for CI/CD, Codespaces, and integrated security scanning. GitLab offers similar capabilities alongside its Duo AI agent platform. Smaller competitors include Supafork, which also markets itself as "AI Agent Native Version Control," and Stoneforge, an open-source project billing itself as "agent-first."
The company has attracted early users from a handful of startups, including Corgi, alphaXiv, AthenaHQ, OpenRelay, and several others, according to logos on GitCafe's homepage. The startup is running trials with roughly a dozen Y Combinator companies, Veronica C. said.
Private repositories are free during the beta. GitCafe's terms of service state the company will publish pricing and provide advance notice before that changes, though no timeline is specified. The privacy policy explicitly says GitCafe does not use repository content to train generative AI models, a notable commitment given how many developer tools companies are racing to incorporate AI features.
GitCafe plans to offer one-command repository migration to ease switching from existing platforms. Diana Hu is the Y Combinator partner working with the company.
A Bet on Velocity
The central wager here isn't just that AI agents will write more code. It's that the velocity and volume of code changes will increase so dramatically that existing infrastructure will buckle. Whether that happens fast enough to justify building an entirely new Git implementation from scratch is the kind of question that separates successful startups from expensive learning experiences. For now, GitCafe is in the early chapters of that story.
