After a repository is connected, an agent studies the codebase and configures a cloud environment for it. That includes databases, infrastructure and sample data, and the setup runs until a live preview works. Each change then runs on an isolated branch. Base44 says the product never pushes to the default branch, so branch protections and required checks still apply. Admins can restrict imports to approved GitHub organizations. Secrets are stored encrypted and are not committed to the repository. Base44 announced the product on September 28, 2026, as a separate offering from its app builder. Until now its GitHub integration synced apps built inside Base44. Base Code instead starts from repositories built elsewhere. Current support documentation limits projects to web or full-stack repositories at the repo root. Users can pick from several AI models, and the workspace is reachable from a browser or phone. Coverage frames the launch as a challenge to Claude Code and Cursor. For AI founders, the notable point is where the agent sits. It is positioned as a way for non-engineers to open pull requests, with engineers keeping merge control. That shifts the buyer from the individual developer to the whole team, and review capacity becomes the constraint. Teams building coding agents should expect rivals to compete on setup automation and governance, not model quality alone.