Wren doesn't just chat. It plans, writes, runs and fixes real code in its own sandboxed workspace, then hands you the result.
Wren takes a task and carries it to a working result, step by step.
Breaks your ask into concrete steps, then works through them in order instead of dumping code and hoping.
Complete, runnable files land in its workspace. Websites, scripts, algorithms — you can open and edit every one.
JavaScript executes in a secure browser sandbox, Python on a real Pyodide runtime. Output comes back to Wren, not just to you.
Tool errors feed straight back into the loop, so Wren reads the failure, patches the code and re-runs until it works.
Ask for answers and explanations. Build for a full agent loop that acts on your workspace files, up to 5 steps per turn.
Everything runs in your browser: the workspace persists locally, and streaming starts the moment you hit send.
"Build a snake game", "why is this regex wrong", "compute prime factorizations up to 10k". Plain language, any size.
It writes files, executes code and reads its own tool output, looping until the result is verified or the answer is solid.
Working files, live output and a summary. Copy anything with one click, or keep iterating in the same session.
One box, two modes, no sign-up. The fastest way to understand it is to watch it work.
Open the playground