Run a search. Watch it pack.
Pick a scaffold — OpenEvolve, Top-K, or Best-of-N — point it at the circle-packing task with your own OpenRouter model and key, and press Run. The best-so-far curve climbs live, and every time the search beats its best valid packing the geometry is captured so you can scrub through how 26 circles rearrange to fill the square.
Configure
idleMAP-Elites islands + adaptive explore/exploit selection (the open AlphaEvolve).
Sent only for this run; used in-process, never stored or logged.
Enter a model and an API key.
Run a search to stream the per-iteration log here.
Each time the search beats its best valid packing, a frame is captured here.
Configure
Choose a scaffold, type an OpenRouter model id and your key, and set how many iterations to run — up to 100.
Run
Your chosen scaffold drives a real evolutionary loop on the bundled circle-packing task. Each candidate is scored in an isolated subprocess; results stream back as they land.
Watch it evolve
The best-so-far curve climbs toward the AlphaEvolve target. Scrub or replay the captured frames to see the packing rearrange — and open the solution code behind any frame.