Can I use Claude Code and OpenAI Codex while still editing my app visually?

Yes. Claude Code and Codex run inside your app’s sandbox and edit the same source files the visual editor renders, so there is no export or import step between them. Prompt an agent, review its changes, then open the screen on the canvas and keep adjusting spacing or copy by hand. Switch direction whenever you like. You might ask Codex to connect an API, tune the layout visually, then start a Claude Code thread to add error handling.

You are not tied to one agent. Every thread chooses its own agent and model, and you can set defaults in Agent Customization. Each agent reads its own instructions file, CLAUDE.md for Claude Code and AGENTS.md for Codex, which is the place to record conventions such as which component library to use.

Plan matters for the hand-editing side. Every plan includes both agents and the visual editor. Saving manual code edits, exporting the code, and opening the sandbox in VS Code, Cursor, or Windsurf require Standard or higher. On Free you can view a selected component’s code from the Code tab, but the full editor and saved edits are paid features. Syncing to GitHub requires Pro.

Highly custom code is the one place the round trip narrows. A component an agent writes may expose only some of its settings in the panels. It still runs, and you can edit it in the Code tab or ask for a revision.

Review the current project before each broad agent task so it does not build on a stale assumption, and save a point before letting an agent refactor widely.