You work on a shared canvas with AI agents — a creative director, image specialists, a critic — and steer them through reactions, selections, and spatial arrangement instead of writing prompts.

It started as “Claude Code, but for media.” Coding agents have compilers and git. Media has none of that. So instead of porting the code agent playbook, we built around what works for creative work: fluid feedback, persistent taste learning, and agents that observe a shared workspace like colleagues.

The engine is use-case agnostic. Hatsets — team blueprints in YAML — configure what agents exist and how they behave. Same engine makes manga, brand ads, album art, interior design mockups.