# OmO

OmO is a single command-line agent, built on a fork of the pi harness, that turns a described job into finished research, presentations, backends, frontends or code by coordinating many specialized agents and models.

Adding a keyword such as "ultrawork" to a prompt makes the agent read the relevant code before touching it, plan the work, prove each step before moving on, and check the result on the real surface rather than stopping early; a second keyword fans a large job out into a graph of many agents running in parallel, each matched to a suitable model. A lightweight companion agent keeps a memory of past work as a git repository of Markdown files and nudges the main agent before it repeats a mistake.

A single login covers subscriptions to several model providers at once, including Claude, ChatGPT, Kimi and GLM, and the project ships built-in skills, including browser use, that load only when a task needs them. It can import settings from an earlier OpenCode-based edition or from LazyCodex.

TypeScript.

Website: https://omo.dev