Some output deserves to outlive the conversation that produced it — a report, a diagram, a small tool, a page you want to show someone. Those become artefacts: saved, listed, and openable long after the chat has scrolled away.
- Artefacts are listed in their own screen and open in a proper viewer rather than as chat text.
- Runnable artefacts can be run again later without re-asking the agent to rebuild them.
- They belong to your account, so an artefact made on the phone is there on the Mac.
Every model, the same artefacts
Artefacts live with your account rather than inside the agent that made them, and every model addresses them by the same identifiers under the same permissions. One built by Claude Code is one Codex can run tomorrow.
That is deliberate. It is what makes changing provider a judgement about which is better at the task in front of you, rather than a migration you keep putting off.
What an artefact can be
- A calculator
- Typed inputs, an ordered set of calculations and a presented result. The same definition drives the web and native controls.
- A page or small app
- Real HTML, CSS and JavaScript with server-side actions, running in an isolated engine. It can call other artefacts you have given it permission to use.
- A browser extension
- Code that runs on the websites you nominate, installed into Ripul’s browser. See Extensions your agent writes.