openai

Codex CLI 0.132.0: Python Auth, Faster TUI, Structured Resume

Codex CLI 0.132.0 brings first-class Python authentication, faster TUI startup with batched terminal probes, smarter goal loops that respect usage limits, and structured output support for resumed automations.

TL;DR

Python SDK now has first-class auth (API key, ChatGPT browser, device-code flows)
TUI startup is faster with batched terminal probes
Goal loops stop at usage limits instead of burning tokens

New

Python SDK authentication — API key login, ChatGPT browser and device-code flows, account inspection, and logout APIs built in. No more workarounds.
Simpler Python turn APIs — Pass plain strings as input; handle-based runs now return richer TurnResult with collected items, timing, and usage data.
codex exec resume --output-schema — Resumed automations keep session context while enforcing structured JSON output.
Faster TUI startup — Terminal capability probes are now batched instead of serial, so the first interactive frame appears quicker.
Standard auth for remote executors — Remote executor registration now uses Codex auth instead of a separate registry credential flow.
Image fidelity preservation — App-server turns keep original-resolution local images across user inputs and image-producing tools.

Fixed

Goal continuation loops — Now stop when hitting usage limits or repeated blockers instead of looping and burning more tokens.
Session picker UX — Renamed threads show name (thread-id) in resume hints; pasted text now works in picker search.
Multi-session TUI reliability — In-progress MCP calls stay marked as active during replay; elicitation replies route back to the requesting thread.
Remote session stability — Websocket connections stay alive; diff paths show repo-relative paths instead of /tmp/... prefixes.
Windows robustness — codex doctor detects npm-managed installs correctly; MSVC release binaries no longer require separately installed VC++ runtime DLLs.
TUI polish — Immediate shutdown feedback on exit, ChatGPT usage link hidden for non-OpenAI providers, cleared Fast tier no longer reappears after side-thread resume.