Analysis updated 2026-05-18
Run existing Claude Code workflow scripts on an OpenAI Codex subscription without modifying them.
Fan a task out to multiple parallel AI agent reviewers and combine their verified results.
Loop a workflow automatically until a skeptical verifier agent approves the output.
| yuanpingsong/ultracodex | n8n-io/tournament | bike4mind/bike4mind | |
|---|---|---|---|
| Stars | 62 | 62 | 61 |
| Language | TypeScript | TypeScript | TypeScript |
| Last pushed | — | 2026-04-28 | — |
| Maintenance | — | Maintained | — |
| Setup difficulty | moderate | easy | moderate |
| Complexity | 4/5 | 2/5 | 4/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires Node.js 20+, the Codex CLI installed and authenticated, and typically a driving agent like Claude Code.
ultracodex lets you take workflow scripts written for Claude Code and run them, unchanged, using an OpenAI Codex subscription instead, and it also supports a third backend called OpenCode. The core idea is that an AI agent can be treated like a unit of programming: you write ordinary JavaScript, call an agent the way you would call a function, and get back a structured result. Because ultracodex hides the differences between the three backends, the same script can run on whichever one you happen to have available, which means the heavy work of running many agent calls can land on a subscription you are not trying to conserve, rather than the one you use for everyday coding. A workflow script is a plain JavaScript module built around a handful of built-in functions: agent, which runs a single AI agent call and can return a schema-validated result or fail gracefully, parallel, which runs several agent calls at once and waits for all of them, pipeline, which pushes items through a chain of stages without waiting for every item to finish the same stage together, plus helpers for grouping progress into phases, logging messages, reading arguments, and tracking a token budget. Beyond running a single workflow, ultracodex adds ways to loop a workflow until a skeptical verifier approves the result, schedule workflows to run on a timer, and, in an experimental mode, keep a longer running group of agents with memory between runs. Getting started involves installing the ultracodex command line tool with npm, running a doctor command that checks your setup, and running a sync command that teaches a coding assistant like Claude Code how to write and run these scripts for you. From there, you can simply describe a task in plain language to your assistant and have it write and execute the workflow, or you can run a script or a named workflow directly from the command line yourself, with options to set a token budget, watch progress live, or run in the background as a separate process. The project requires Node.js version 20 or newer and the Codex command line tool to be installed and logged in. It is released under the Apache License 2.0.
ultracodex runs Claude Code style agent workflow scripts on an OpenAI Codex or OpenCode subscription, letting you fan out, loop, and schedule AI agent calls from plain JavaScript.
Mainly TypeScript. The stack also includes TypeScript, Node.js, Codex CLI.
Apache License 2.0: use, modify, and distribute freely, including for commercial purposes, with attribution and a patent grant from contributors.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.