OpenAI has released a new version of its Codex coding tool. It preserves drafts when switching connected servers in the terminal and keeps saved conversations accessible after model provider changes. This makes it easier for developers to continue earlier work across environments.
1. Keep drafts across connected backends
This version lets users switch backends, the connection targets that handle work, in the terminal session picker. Sessions are connection units that retain work state. Selecting an already connected backend preserves the draft during the switch. Users can change connections on the same screen without copying unfinished input elsewhere.
Saved conversations now remain listed after a model provider change. The update preserves drafts when switching connection targets and access to saved conversations after changing providers.
Some saved conversations cannot be resumed directly. In those cases, users can read the history and prepare a prompt with requests and instructions telling the model what to carry forward into a new conversation.
2. Find unread results and conversation history more easily
The command center, which gathers work by agents that perform tasks on users' behalf, is now available in sessions embedded in other screens. The command center and session picker indicate unread results or sessions. Opening the history marks that session as read. This helps users identify unread items when checking results from multiple tasks.
Open the command center with this command.
codex agents --no-daemonCtrl+L, which clears the terminal screen, has also changed. It clears existing screen content but preserves previous messages, which can be viewed again with Ctrl+T. Separating screen cleanup from history deletion keeps earlier messages accessible while tidying the terminal.
3. Refine image and change review workflows
Supported terminals can now preview local image attachments while users compose a message. Previously viewed local images can be reopened by clicking their paths in the conversation history. This simplifies moving between attachments and earlier records in tasks involving images.
Users can also expand patch details, which show file changes, in the conversation history while an approval request is open. This lets them assess approval and review patches in the same conversation screen without leaving to inspect changes.
4. Organize sign-in choices and proxy settings
API key and cloud provider options are now grouped under the 'Sign in another way' submenu. The Google Cloud Vertex AI setup wizard can configure a connection using Application Default Credentials or a credentials file to verify identity.
System proxy settings on macOS and Windows can be enabled from the experimental settings screen. A proxy is an intermediary server that forwards network requests. Users who want to use their operating system's proxy settings must enable this option themselves.
The wait time for OAuth sign-in, which authenticates through a provider account, has increased to 10 minutes. This gives users more time to complete the sign-in process.
5. Continuity benefits and configuration overhead
Preserving history and drafts and marking unread results reduces the effort of finding earlier content when switching backends or tasks. However, broader options such as cloud provider authentication and system proxies require users to choose authentication methods and connection settings suited to their environments.
The release notes cover managed network restrictions that let administrators limit network access on the app server. They also cover requirements for fresh human approval each time under prompt-user command rules. Environments using these approval rules prioritize human checks over the convenience of automatic progress.
Network approval dialogs now close when their requests finish, and concurrent requests to the same destination ask for approval one at a time. Users who need connection controls should review both authentication settings and the workflow for handling approval requests.
Codex rust-v0.167.0 adds draft and conversation history continuity to the flexibility of switching backends.
