Skip to content
NewsProduct launch

GitHub open-sources Replica, an AI prototyping tool built from design systems

· by Pondero Newsdesk

The short version

GitHub's design team open-sourced Replica on October 28, 2026, an MIT-licensed tool that builds AI prototypes from a team's own components instead of generic UI, and the repo has already drawn outside contributions.

GitHub open-sources Replica, an AI prototyping tool built from design systems

GitHub's design team released Replica as a free, MIT-licensed repository on October 28, 2026, and the pitch is narrower than the usual AI-prototyping tool: it generates interactive mockups constrained to a team's own design system, not whatever components a model decides to invent, per GitHub's changelog post.

What shipped

Replica works inside the GitHub Copilot app. A designer first defines guidance drawn from the team's existing components, patterns, and documentation; anyone else on the product team can then describe a feature in natural language and get back a working prototype that reuses those same pieces, per GitHub. The repository ships three working parts: a Replica plugin that opens a live preview canvas next to the Copilot conversation, an optional self-hosted "Replica Hub" for publishing prototypes and collecting comments, and a GitHub Actions build step that packages a prototype for Hub. New projects start from one of two starter kits, a bare-bones minimal app or a "CoreAI Sandbox" prebuilt on Microsoft's CoreAI design system, according to the repo's README. Model access runs through a GitHub Copilot plan by default, or through a bring-your-own-model option that GitHub labels public preview and subject to change.

Running Replica locally needs only the Copilot app, Git, and Node.js on macOS or Linux, no server required. Hosting a shared Hub is the heavier lift: a dedicated Linux server, a wildcard-DNS domain with HTTPS, and GitHub permissions approvals, per the repo's setup docs. Five days after launch, the repository had drawn 132 stars, 17 forks, and 72 commits from at least three contributors, including an automated dependency-update bot, per GitHub's repo page, a small but real signal of outside engagement for a design-tooling release.

The constraint, not the generation, is the product

Tools that turn a prompt into a working UI are common now; what Replica is actually selling is the opposite of that freedom. By forcing every generated screen through a team's existing component library and approved patterns, GitHub is betting that prototypes a product team can trust enough to user-test matter more than prototypes that look polished but drift from what engineering will actually ship. That is also why the project ships as a plugin GitHub itself can run internally rather than a hosted SaaS app: a design system is proprietary by nature, and self-hosting the Hub keeps a company's component library and prototype traffic off a third party's servers. Teams evaluating AI prototyping tools now have a concrete, free option built specifically for that constraint, rather than for raw generation speed.

What to watch next

Worth tracking is whether Replica gets meaningful adoption outside GitHub's own design org, given it launched without an accompanying case study naming an external customer, and whether the bring-your-own-model option moves out of public preview. The open issue tracker and pull-request queue, both public on the repo, will show whether outside teams are filing real feature requests or just forking it quietly.

Sources