OpenAI Upgrades Codex CLI to Run Parallel Coding Agents

OpenAI has released a major update to Codex CLI, its open-source terminal-based coding assistant. The refreshed tool introduces a full-screen terminal user interface designed specifically for managing long-running, concurrent developer workflows. The new layout features a pinned prompt composer at the bottom of the screen, an on-demand scrollback history that bypasses standard terminal limits, and collapsible diffs and tool outputs. It also renders Mermaid diagrams and LaTeX formulas directly inside the terminal window.

The update focuses heavily on multitasking through several new slash commands. The /agents command displays a central dashboard of all active tasks, allowing developers to monitor progress and switch between parallel Codex sessions. To prevent overlapping code changes, the /fork command now automatically clones a conversation and maps it to a separate Git worktree. This allows developers to test competing implementations in isolated directories while sharing the underlying repository data.

Additionally, the /voice command introduces two-way spoken communication, enabling developers to speak instructions and hear audio responses directly in the terminal using F8 as a toggle. OpenAI has bundled the necessary audio runtimes for both Linux and Windows to simplify setup. Other additions include /usage for tracking session activity and /theme for changing visual styles. The tool remains open source under the Apache 2.0 license and can be installed or updated globally using the command npm install -g @openai/codex@latest.

For software engineers, this release shifts the terminal from a simple chat interface into a comprehensive control center for supervising multiple AI agents. By integrating Git worktrees directly with conversation forks, the tool reduces the friction of managing parallel feature branches. However, because developers can now trigger multiple concurrent workflows, the primary bottleneck shifts from writing code to verifying and merging the resulting diffs. Users can authenticate using either a ChatGPT account or an API key, with billing tied to their respective plans or API project usage.