Phasis develops platform for agentic orientation using World APIs and reality layers

Phasis is trying to give AI agents a better sense of where they are before they start making moves. The company’s bet is that graphs, cleaned evidence, and ontologies can turn messy real-world context into something agents can actually use.

Brandon Farley, Phasis’ co-founder and CEO, said the work started with culture: not just the ties between people, but the ties between people and their values, beliefs, and subcultures. That led the team to the same basic method across other domains — clean the data, verify it, build the ontology, and represent reality in a form an agent can orient against.

That is the pitch behind the company’s World API. A World is a maintained, evidence-backed state of a domain, and Phasis builds and deploys those Worlds so agents can begin with current context and keep up as things change. The company says the aim is to create a reality layer for production agents, one that can be shared or private depending on the use case.

The main entry point is an endpoint called /orient. Send it a task, and Phasis returns the relevant World frames: current state, material changes, supporting evidence, and uncertainties. Farley described it as a quick check against reality before an agent searches, plans, or strategizes. The point is not to replace those steps, but to stop agents from stumbling in blind.

Farley also argued that the advantage compounds over time. In his view, the companies that keep their agents oriented against reality will pull ahead of the ones whose systems have to piece things together from scratch every time. He said smaller models are already capable, and he expects the next year to bring more agent swarms and auto research.