Overview

OmO is a command‑line agent that turns a single keyword into an orchestrated flow of tasks. By typing mass ulw in a terminal, you launch a graph of agents that can browse thousands of sources, train ranking models, generate investor decks, and more. The tool is available for macOS, Linux, and Windows via PowerShell or CMD and installs with a short shell script.

The agent works by composing specialized models, Claude, GPT, Kimi, Grok, GLM, DeepSeek, into a pipeline. Each task is assigned to the model that best fits its needs: high‑speed coding with Kimi, visual engineering with GPT 6 Astra, or hypothesis testing with Grok 4.6. The agents run in parallel, verify results against each other, and report progress in a step‑by‑step workflow. If a task feels difficult, adding the word ultrawork prompts the agent to read the whole codebase, plan each step, and validate outputs against real data before proceeding. Typical input for a user is a natural‑language prompt such as “make the quarterly report actually readable ultrawork.” The output is a verified report, a deck, or a code change that has been checked against the local repository.

OmO also includes a lightweight assistant called Kibitzer. A small model watches the main frontier agent, remembers prior interactions, and nudges it to avoid repeating mistakes. This two‑tier approach keeps expensive calls to frontier models minimal while maintaining quality. The software ships with a library of pre‑built skills, frontend design, visual QA debugging, data science workflows with DuckDB and Polars, that the agent loads only when needed. For example, a user can request a chart from a raw CSV; OmO will launch a data‑science skill that reads the file, runs Polars queries, and outputs a PNG chart.

Because OmO is script‑driven, you can integrate it into existing CI/CD pipelines or run it directly from your shell. The terminal displays a progress tree for each wave of the workflow, listing pending sub‑tasks such as “source‑sweep” or “gate‑review.” The tool also offers a manifesto page with documentation on workflow terms like “mass ulw,” “ultrawork,” and the various model names. The listing does not disclose pricing, detailed error handling, or limits on the number of concurrent agents beyond the implicit limits of the underlying models.