Codex CLI Ships Three Updates in a Week: Web Search in Code Mode, Claude Code Imports, Token Tracking

OpenAI shipped three Codex CLI releases in quick succession between June 8 and June 15. Taken individually, most of the changes are incremental. Together, they add up to a noticeably more capable terminal agent, particularly for developers who split time between CLI and desktop workflows.

0.138.0 (June 8): Hand Off Sessions to the Desktop App

The 0.138.0 release added the /app command, which hands the current CLI thread off to Codex Desktop on macOS and Windows. If you started a session in the terminal, built up context, and now want to continue in the desktop app’s GUI, /app carries the conversation over. No copy-pasting, no starting fresh.

The same release also changed how local image attachments are handled. Image files now expose their saved file paths to the model, which makes follow-up edits more reliable. If you attach a screenshot and ask Codex to reference it in a later turn, it can find the file instead of treating the attachment as ephemeral.

0.139.0 (June 9): Web Search in Code Mode, Computer Use on Windows

Code mode, which gives Codex direct access to your file system and tools without an interactive UI, can now call web search directly. That includes from nested JavaScript tool calls, which matters if you’re building agents that themselves invoke tools and need live web data mid-execution. Results come back as plaintext.

Computer Use support landed on Windows in this release. It was previously macOS-only for desktop control. Windows users can now run Codex sessions that control apps, fill forms, and navigate interfaces without the limitation.

The release also fixed a bug in MCP tool schema handling. Tool/connector input schemas now preserve oneOf and allOf constructs properly, which was causing problems with complex schemas from MCP servers. If you’ve been working around mangled schemas, 0.139.0 should resolve it.

0.140.0 (June 15): Import From Claude Code, Token Usage View

The most recent release has two features worth paying attention to.

/import: Bring Your Claude Code Setup Into Codex

Codex now has a /import command that selectively imports setup and project configuration from Claude Code. During onboarding and from the Settings screen, you can choose which parts to bring over: skills, hooks, MCP servers, subagents, instruction files, and up to 30 days of session history.

OpenAI also added formal Migrate to Codex flows in the desktop app, accessible via Settings > Import other agent setup. The importer scans your machine for Claude Code configurations and migrates what it can handle automatically, then opens a follow-up thread for anything that needs manual attention.

Whether you read this as “feature for Claude Code users considering switching” or “quality-of-life for developers who run multiple agents,” the importer is a concrete acknowledgment that developers often work across tools rather than committing to one.

/usage: Token and Spend Tracking

/usage is a new slash command that shows daily, weekly, and cumulative account token activity broken down by model. If you’ve been trying to understand how much your Codex sessions cost or which sessions are driving the most consumption, this gives you that data without leaving the terminal.

The release also unified the mentions system: typing @ in the composer now opens a single menu for files, plugins, and skills, rather than separate pickers depending on what you’re trying to insert.

Amazon Bedrock managed API key authentication with encrypted local storage was added for users connecting Codex to Bedrock endpoints. Credentials are cached with managed rotation rather than sitting in environment variables.

Update

Run npm update -g @openai/codex or codex update depending on how you installed. The full changelog is at developers.openai.com/codex/changelog.

Sources: