Image: OpenAI
Why it mattersTeams running Codex on cloud providers get a supported path to Vertex AI. Long-running agents are now easier to watch.
OpenAI released Codex version 0.167.0 on 6 November. The release notes list new features for people running long-lived coding agents, and for teams on Google Cloud. The version is the fourth since 0.163 was released on 16 October.
The agent command center
Codex now opens an agent command center from the terminal with the flag codex agents --no-daemon. The command center and the session picker mark results as unread, OpenAI writes. A user can see which running sessions have finished since they last looked. The user does not have to open each session in turn.
The release lets a user switch between connected backends from the terminal session picker. Drafts stay in place when the user switches. Local image attachments can be previewed inside the composer on supported terminals. Images open from clickable transcript paths.
Google Cloud Vertex AI setup wizard
The release adds a setup wizard for running Codex against Google Cloud Vertex AI. From the terminal, gcp_setup_wizard sets up Application Default Credentials or a credential file, OpenAI writes. Vertex AI and other cloud provider options now live under a "Sign in another way" entry during sign-in. The step before this required a user to configure Vertex AI by hand.
Users can now opt into the macOS or Windows system proxy settings from the experimental menu. Earlier versions did not offer this.
Bug fixes and Linux memory
Codex saved chats now stay visible after a user switches model providers, OpenAI writes. If a chat cannot resume, the transcript stays readable. The user can prepare an editable prompt to continue in a new chat. Ctrl+L clears the terminal display while keeping earlier messages available through Ctrl+T.
Network approval prompts close when their requests end. Concurrent requests to the same endpoint prompt one at a time. Long realtime V3 conversations now save completed speech to chat history as the transcript grows. Automatic context compaction now accounts for retained output from failed responses, and responses with missing token usage data.
Codex's Linux musl builds use fewer allocator arenas. That reduces idle memory after conversations unload, OpenAI writes. The release also wires Code Mode into the runtime. It adds a bounded output collector, server-side truncation, and a per-chat lease for shared MCP connections.
The release is tagged rust-v0.167.0. Users migrating off a deprecated model now get a hint through the openai-docs skill. The hint names the documented replacement.
The release notes also list smaller changes for people writing their own skills. The app-server README now covers managed network constraints. It documents prompt-user command rules that require fresh human approval. Patch details now expand directly in the transcript while an approval prompt is open. A pasted search text in the skills toggle view is now supported.
The changelog lists more than 100 pull requests. OpenAI has released four Codex versions in the three weeks since 0.163.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
