Omnigent setup: Claude Code implements, Codex reviews

Omnigent is presented as an open source orchestration layer that runs on your machine and coordinates multiple coding tools. The tutorial covers setup through its repository and explains how it reuses credentials from CLIs you have already signed into. The speaker reports getting a Claude Code and Codex workflow running in less than ten minutes.

The main example uses Polly to delegate implementation to Claude Code and send the result to Codex for review. The changes remain in a local Git worktree for follow-up work. The speaker recommends separate implementation and review sessions, while acknowledging that preferences for particular models are not universal rules.

Custom orchestrators combine configuration, workflow skills and agents they can call. The walkthrough covers system prompts, tool access and sandbox options, including Docker and E2B. A custom Python policy pauses forced Git pushes for human approval. Another example, Debbie, runs a debate between Claude and GPT, exchanges their feedback and synthesizes an answer.

For local AI workflows, the speaker describes using Pi with Ollama, though the main demonstration uses Claude Code and Codex. The tutorial also demonstrates a phone message appearing in the same desktop session. It describes shared Wi-Fi as the easiest device setup and points to repository instructions for server deployment and access across the internet.