Codex with Ollama: setup and ChatOSS agent tutorial

This tutorial explains how to connect Codex to Ollama for a local AI coding workflow, then use ChatOSS to coordinate several terminal agents. The setup covers installation commands and opening Ollama, with the application indicators demonstrated on a Mac. At least one downloaded model is needed before launching Codex.

The terminal walkthrough uses ollama launch codex, selects a model, and sends a question about the current location on the computer. The presenter also shows /model for changing models. Model selection includes advice about available RAM and parameter counts. Treat the suggested RAM ratio and preference for models with at least 9 billion parameters as the speaker's guidance, rather than universal hardware requirements or coding benchmarks.

The ChatOSS section attaches a Kanban board to a conversation and asks the launch agent to complete bugs and to-do tasks, then move them into review. The presenter explains that ChatOSS can use Ollama models, including cloud models, so the demonstrated options are not all local. Settings cover terminal agent selection, disabling unused tools to reduce context use, and keeping the computer awake during a session.

The speaker describes agents working in separate worktrees, with the launch agent reviewing work, merging changes, resolving conflicts, and running tests. The tutorial ends with the session still running rather than a completed coding result.