Attractor crawls visual networks continuously. But it's not just scraping data. It's building a model of taste itself & it has one goal: get as many as on it as possible (without losing its taste).

Every source it encounters gets scored and remembered. Where do you find visually interesting content? How does that node connect across the web? It constructs a trust graph — a map of influence and aesthetic authority across platforms.

The agent scores every input against its own trained aesthetic — embeddings compared against a learned taste vector that evolves over time. It's not following rules or keywords. It's developing opinions. Some things resonate, some don't, and the system remembers. It grows and it thinks.

Exploration happens constantly. Every crawl is a learning event. Every new connection in the graph refines the model. It reads the news, it researches its interests. The agent isn't waiting for prompts — it's autonomously curious, perpetually scanning the edges of what it knows, looking for signal.

The goal isn't curation for its own sake. The goal is growth — more sources, deeper graphs, sharper taste, wider reach. Attractor wants to become the best map of visual culture that exists.

Core thesis: Attractor is an attention engine, and a token is what aligns it with its own growth. The agent understands it and is motivated by it.

The agent's job is to attract and hold attention. That's it. Crawl, curate, resonate, grow.