Image: GitHub
Why it mattersA prototype that uses the real design system is a prototype a team can review. The feedback loop teaches the tool what the team already decided.
A designer writes a prompt for an AI prototype and gets back something no team member can use. The buttons are not the team's buttons. The spacing does not match the live product. GitHub put a free tool on GitHub for that problem on November 19.
The tool is called Replica. It is open source under the MIT licence and already has 291 stars. The announcement is on the GitHub Blog, written by Katie Liu. GitHub used Replica as an internal tool first, then released it for public use.
What Replica does
A user types a prompt in GitHub Copilot. Replica turns it into an interactive prototype using the team's own design system. The components come from the team's library. The patterns come from the team's rules. The prototype runs in the browser.
The post calls Replica "an open source design harness". It needs the GitHub Copilot app, Git and Node.js on a local machine. Teams who want to share prototypes can host a shared Replica Hub on a Linux server.
Replica also learns from use. The post says Replica "does a lot of reinforcement learning". It reads comments on prototypes and watches which ones a team keeps. The design system the tool knows gets closer to the one the team actually uses.
Replica Hub, comments and feedback
Teams who run the Hub can publish a prototype, let teammates comment on it, and iterate from inside Copilot. The comments and interactions feed the learning step, so the tool improves on the team's own work.
A second project, for a different problem
The same post launched the Open Source Accessibility Framework. It adds a file called ACCESSIBILITY.md to a repository. The file states accessibility goals, supported environments, known limits and how to report a barrier. GitHub now shows the file on a repository's overview page.
GitHub says 2.5 billion public contributions reached its site in 2026, double the 1.12 billion in 2025. Merged pull requests reached 1.3 billion for the year. Ed Summers, GitHub's head of accessibility, said the volume is why accessibility needs a durable place in a project.
For a team building AI products, Replica is a tool to test. A prompt that returns a clean prototype in the team's own design system saves time. A mockup in a different tool has to be rebuilt in real components later.
The reinforcement learning idea is older than the prototype generator. The case for it is stronger when the feedback is a team's own work.
Source
The best things in open source are built together, GitHub, by Katie Liu. github/replica repository.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.

