OpenAI Codex Interaction Surfaces Overview and Architecture

Overview

OpenAI Codex delivers model capabilities across four distinct interaction surfaces: CLI, Desktop App, Web, and IDE extensions. All surfaces operate under a single account and unified usage allowance, making surface selection an architectural decision based on workflow preference, local environment constraints, and task parallelism.

Key Insights

• Shared Account Model: A single paid ChatGPT subscription grants access to all four surfaces with a centralized usage allowance, eliminating vendor lock-in to any specific UI.
• Consistent Local Security: The CLI and Desktop App utilize the same open-source, configurable, system-level sandboxing for local environment isolation.
• Parallel Orchestration: The Desktop App and Web interfaces streamline asynchronous multi-agent orchestration, whereas the CLI optimizes for single-repo, low-overhead local operations.
• Remote Offloading: Web and IDE interfaces support cloud-side execution, allowing developer machines to remain unencumbered by heavy computational background tasks.

Technical Details

Interface Architectures

Codex CLI

• Execution Environment: Runs locally inside the terminal shell; open-source under the Apache 2.0 license.
• Capabilities: Supports interactive REPL modes and direct one-shot script commands. Provides maximum control and lowest UI latency for terminal-first developers.
• Trade-offs: Running parallel agent threads requires manual management of multiple terminal windows, shell sessions, and Git worktrees.

Codex Desktop App

• Execution Environment: Runs locally on macOS and Windows using native system-level sandboxing.
• Capabilities: Optimized for supervising multi-threaded agent workflows. Manages discrete threads, worktrees, and review contexts within a unified GUI.
• Trade-offs: Lacks native Linux support; adds visual overhead for rapid single-file or single-repo modifications.

Codex Web (chatgpt.com/codex)

• Execution Environment: Runs entirely within OpenAI’s remote cloud infrastructure.
• Capabilities: Requires zero local setup. Executes asynchronous background tasks against remote sandboxed git checkouts, accessible from any browser.
• Trade-offs: Operates on remote checkouts rather than direct local workspace files, requiring explicit synchronization.

IDE Extensions

• Execution Environment: Embedded directly into VS Code, Cursor, Windsurf, and JetBrains environments.
• Capabilities: Offloads heavy processing jobs to cloud sandboxes while streaming resulting code diffs directly into the local editor context for inline approval.
• Trade-offs: Tied to editor lifecycle and window management.

Workflow Selection Matrix