Image: OpenAI / github.com/openai/codex Codex CLI 0.148 and 0.149: Session Forking, /export, and an Agents Dashboard
OpenAI shipped two Codex CLI releases this week. Version 0.148 adds session forking and Markdown export. Version 0.149 adds an interactive agents dashboard and working directory commands.
OpenAI shipped two Codex CLI stable releases this week: 0.148.0 on August 18 and 0.149.0 on August 20. Together they add session branching, conversation export, an interactive task dashboard, and better working directory control from inside a session.
0.148.0: Session Forking and Export
Session forking is the headline feature. Running codex exec fork branches a session from its current state rather than replaying it from the beginning. If you’re mid-session and want to explore two different approaches without losing your current context, you fork and diverge. The resume picker now shows archived sessions alongside active ones, so you can jump between branches without leaving the TUI.
/export writes the complete conversation from a TUI session to Markdown, either to the clipboard or to a file. This is useful for logging what Codex did in a long agentic run, filing a summary in a ticket, or reviewing a session later without re-opening it.
Cost and usage tracking in /status now shows estimated thread credits or cost for eligible workspaces. The same number appears in status lines and terminal titles. For teams tracking AI spend at the session level, this moves the number somewhere visible without requiring a separate dashboard.
Amazon Bedrock Runtime is now a built-in provider. Set your AWS profile and region, point Codex at your Bedrock endpoint, and the authentication and model routing is handled natively.
Hook improvements: hooks can now run commands asynchronously and invoke MCP tools. Previously, hooks were fire-and-forget shell commands. The ability to call MCP tools from a hook opens up integration patterns that weren’t possible before.
On the stability side: model switches no longer leave stale system prompts in an active turn, turns reconnect through temporary provider outages, and MCP servers recover after OAuth reauthentication without needing a restart.
0.149.0: Agents Dashboard and Directory Commands
codex agents is a new interactive dashboard for managing running and paused tasks. From the dashboard you can search, start, open, rename, and stop tasks. Keyboard shortcuts are configurable. If you’re running multiple parallel Codex sessions, this gives you a single place to check status and context-switch rather than flipping between terminal windows.
/cd, /pwd, /cwd add working directory management to TUI sessions. Previously, if you started a session in the wrong directory or needed to switch mid-session, your options were limited. These commands let you navigate without exiting.
codex queue lets you send a message to an existing local or remote session without opening it interactively. Useful for sending follow-up instructions to a long-running agent task that’s running headless.
Vim editing got expanded: character replacement and additional change motions (cw, c$, cc) now work in the composer.
codex doctor now checks for endpoint protection software, network and proxy failures, desktop app state, and update connectivity issues. The diagnostics were previously focused on authentication and environment config. The expanded checks should catch more setup problems before they show up as confusing errors mid-session.
The TypeScript SDK also gained the ability to pass exact CLI configuration overrides and select max or ultra reasoning effort levels programmatically.
Getting It
npm update -g @openai/codex