The Codex Changelog

Learn how PR 59447 restores the features.daemon_auto_start flag, giving developers and admins control over persistent background processes in Codex. We cover how to configure the opt-out, clean up leftover processes, and review additional updates to Guardian context budgeting and Windows sandboxing.

Codex 0.166.0 introduces experimental connection sharing for local stdio MCP servers to eliminate duplicate processes, reduce memory bloat, and isolate request timeouts across chats. We explore how to configure the feature flag, navigate potential initialization panics with legacy servers, and review smaller fixes like the network-only doctor flag.

Codex CLI 0.166.0 introduces an interactive recovery flow for biological and cybersecurity policy blocks, allowing developers to fork sessions from before blocked turns using selectable safer rewrites. This release also fixes a double-counted reasoning token bug across thread forks, improves TUI Markdown rendering, and updates the npm wrapper to use execve.

In Codex 0.166.0, the default thread unload delay drops from thirty minutes to sixty seconds, creating an operational trap where queued commands can sit idle on unloaded threads. Learn how to use the new --resume flag on codex queue to wake dormant threads and avoid silent automation stalls.

A deep dive into Codex CLI 0.166.0, focusing on the newly packaged standalone codex exec server binary and key packaging gotchas across npm and PyPI. We also cover crucial reliability fixes that preserve permissions, approvals, and reasoning token counts across thread resumes and forks.

Show Notes

• Releases · openai/codex - GitHub: https://github.com/openai/codex/releases

Codex CLI 0.166.0 introduces keyboard-driven transcript selection in the TUI, featuring motions, integrated search, overlay support, and preserved composer drafts. We also cover quality-of-life updates, including the new resume flag for codex queue and network-only diagnostics in codex doctor.

In Codex CLI 0.165.0, Memory V2 becomes the default engine for freshly initialized profiles, enabling local asynchronous context persistence across sessions. We also explore external memory imports from tools like Claude Code and Cursor, session-level controls via /memories, and several quality-of-life CLI fixes.

Codex CLI 0.165.0 adds native support for Google Cloud Vertex AI using Application Default Credentials, bypassing custom proxies. We break down the required environment variables, provider caveats, and smaller fixes for reasoning effort, git worktrees, and marketplace URL validation.

OpenAI has released GPT 6.1 Terra, offering a 50% price cut, faster execution on multi-step terminal tasks, and unlocked sampling parameters when reasoning effort is set to none. Learn how to switch to Terra inline in Codex CLI 0.164.0 and explore key updates in Codex Security 0.7.0.

Learn why transitioning between the Codex CLI and Desktop app can trigger thread writer lock errors caused by the managed app server daemon's 30-minute grace period. We explore workarounds like selective daemon stops, chat forking, and the unconfirmed --no-daemon flag, alongside fixes and bugs from recent 0.164.0 and 0.165.0 alpha releases.