Alloy · AI prototyping with your real product
Alloy is an AI prototyping workspace for changing an existing product rather than starting with an unrelated mockup. The official product page describes browser capture, codebase connection, chat-driven edits and cloud sessions that can be shared for feedback. The choice of starting material affects what the session can do.
Choose capture or codebase access
The browser extension captures pages for visual prototyping. Connecting a codebase is the separate route that lets Alloy run the app and write code. Do not assume a captured page establishes complete access to the underlying application logic.
The publisher says codebase setup requires admin permissions. Establish who may authorize repository access and which project is in scope before connecting it. Start with a bounded design change so that reviewers can distinguish the requested work from unrelated changes.
Alloy also describes libraries and reusable components based on the team's design system. Existing components, tokens and assets can provide a more specific brief than asking an AI to approximate a brand from a screenshot.
Review in the cloud, then review the code
Sessions run in a sandbox with a live browser preview. Shared links and comments let colleagues or customers discuss the proposed interaction without reproducing a local development environment. Inline comments can attach feedback to an element and page.
The product supports converting a session into a GitHub pull request. A pull request is a review handoff, not proof that the change is safe or production-ready. Check the diff, behavior and required tests before merging or releasing it.
Alloy's FAQ says Figma import is not currently available. It separately documents exporting a prototype to Figma. Keep those directions distinct; a label in a flattened pricing table should not override the explicit availability statement.
Match access to the plan
The Free plan lists unlimited sessions and viewers alongside a limited monthly credit allocation. Unlimited sessions therefore do not establish unlimited model use. Pro adds larger credits, top-ups and code-export access; Enterprise lists custom terms and administration features.
The site also describes using an existing Claude or ChatGPT subscription for supported models. That feature does not establish that every model, action or session is covered by an external subscription, so confirm the selected route and its limits.
Review link-sharing permissions before exposing a prototype that contains private product information. This overview describes the published workflow, not a repository connection, account-based prototype test or independent audit of its sandbox.