This changelog tracks Codex releases across the CLI, desktop app, and IDE extension. For the full official changelog, see developers.openai.com/codex/changelog.
Strengthened dangerous-command detection — Codex now recognizes more forced-rm variants and returns clearer messages when it blocks an operation (v0.144.5)
Corrected the bundled instructions for GPT-5.6 Sol, Terra, and Luna so Codex operates them with their true 272,000-token working context (v0.144.6). These models still expose a larger context through the OpenAI API — 272K is the window Codex itself drives them with.
Reliability fixes across v0.144.2–0.144.4
Codex joins the ChatGPT desktop app - July 9, 2026
Codex is now integrated into the ChatGPT desktop app on macOS and Windows; existing Codex projects, settings, and workflows carry over
Edit Markdown and code directly in the app, annotate selections inline, and ask Codex to revise selected content
Review GitHub pull requests from the sidebar with feedback beside the diff, and group multiple repositories in one project
Computer Use is faster with GPT-5.6, plugin management moved into Settings, and task progress plus mobile/SSH reliability improved
CLI v0.144.0–0.144.1 - July 9, 2026
New writes app-approval mode permits declared read-only actions and asks before writes
MCP tools can request authentication interactively without an experimental flag; app-server hosts can provide Codex auth at runtime
Ultra reasoning warns when high multi-agent concurrency may consume usage quickly
Resumed tasks recover from retired models by retrying with the currently selected model
System-proxy-aware Responses WebSockets, improved Windows sandbox support, safer terminal paste handling, and installer/Code Mode reliability fixes
CLI v0.143.0 - July 8, 2026
Remote plugins are enabled by default, with richer catalog metadata and npm marketplace sources
Added codex remote-control pair for manual pairing codes from a running daemon
Amazon Bedrock gained GPT-5.6 Sol, Terra, and Luna plus first-class max reasoning effort
MCP tools use tool search by default; ChatGPT-hosted MCP servers can use session authentication
App-server clients can inspect environments, list descendant tasks, and fork history through a selected turn
ChatGPT for iOS + CLI v0.142.5 - July 1–6, 2026
ChatGPT for iOS can create, search, open, fork, and manage Codex tasks, inspect staged/unstaged and branch diffs, attach files, connect to SSH hosts, and show usage/credit details
Task state, plugin autocomplete, workspace diffs, reconnection, host pairing, and foreground recovery were improved
CLI v0.142.5 stopped full Responses WebSocket request payloads from being written to trace logs
Codex App v26.616 — Record & Replay (June 18, 2026)
Show Codex a workflow once, reuse it as a skill
Record & Replay (macOS) turns a demonstrated workflow — filing an expense report, submitting a time-off request — into an inspectable, editable skill. You control when recording starts and stops.
Remote executors now use authenticated, end-to-end encrypted Noise relay channels, and cross-platform remote execution preserves each executor’s native working directory and shell
Selected executor plugins can activate their stdio MCP servers per thread
App-server clients can list immediate child threads and correlate external-agent imports
Realtime clients can explicitly append speech and control response entry into conversations
TUI input prompts can auto-resolve after inactivity, with a countdown that pauses when you interact
Fixes: hook-trust bypass now persists through codex exec; Windows sandbox repairs stale credentials; bundled SQLite pinned to the WAL-reset fix; TLS now supports P-521 certificate signatures
Performance: tool-search caching cuts latency and memory in large, tool-heavy sessions
CLI v0.137.0 - June 4, 2026
F13-F24 Keybindings, Controller Pairing, Parallel Web Search
Install: npm install -g @openai/codex@0.137.0
TUI adds F13-F24 keybinding support and paste into searchable menus
Enterprise flows surface monthly credit limits and support cloud-managed config bundles for EDU workspaces
codex remote-control clients can initiate pairing and manage controller grants via app-server v2 RPCs
Plugin workflows gain machine-readable JSON output and cached remote catalog suggestions
Hosted web and image tools reach more code-mode flows, with a parallel standalone web-search capability
MultiAgentV2 preserves the runtime choice per thread, with cleaner follow-up and metadata defaults
CLI v0.136.0 - June 1, 2026
/archive Sessions, Clickable Links, Windows Elevated Sandbox Setup
Install: npm install -g @openai/codex@0.136.0
Archive sessions with the /archive slash command or codex archive / codex unarchive; archived sessions are protected from resume and fork
TUI markdown keeps clickable web links via OSC 8 metadata; cramped tables collapse into readable key/value records
App-server integrations resume threads with an initial turns page, plus richer MCP server status visibility
Windows admins gain an alpha codex sandbox setup --elevated provisioning path
ChatGPT auth refreshes tokens before the five-minute expiry window, with better handling of reused refresh tokens
codex doctor Diagnostics, Vim Text Objects, Cleaner Tables
Install: npm install -g @openai/codex@0.135.0
codex doctor now reports richer environment, Git, terminal, app-server, and thread-inventory diagnostics for support cases
/status shows remote connection details and server version when the TUI is connected over a remote transport
Vim mode gains text-object editing, improved word/line-end motions, and a configurable interrupt-turn binding
/permissions understands named permission profiles and lists configured custom profiles
Markdown tables and multiline lists render more readably in the TUI; output is more stable on macOS and Zellij
Python SDK exposes friendly Sandbox presets for thread and turn APIs
CLI v0.134.0 - May 26, 2026
History Search, --profile Standardization, MCP Env Targeting
Install: npm install -g @openai/codex@0.134.0
Search across local conversation history — case-insensitive content matches with result previews
--profile is now the primary profile selector across the CLI, TUI permissions, and sandbox flows; legacy profile configs are rejected with migration guidance
MCP setup gains per-server environment targeting and OAuth options for streamable HTTP servers
Read-only MCP tools can run concurrently when they advertise readOnlyHint
Richer extension/hook context — conversation history for extension tools, subagent identity in hook inputs
CLI v0.133.0 - May 21, 2026
Goals On by Default, codex remote-control Foreground, Permission Profiles
Install: npm install -g @openai/codex@0.133.0
Goals are now enabled by default, backed by dedicated storage and tracking progress across active turns
codex remote-control runs like a foreground command — waits for readiness, reports machine status, and keeps explicit start/stop daemon commands
Permission profiles gain list APIs, inheritance, managed requirements.toml support, and stronger Windows sandbox integration
Plugin discovery is easier to inspect: marketplace-aware list output, installed versions, visible marketplace roots, remote collections
Extensions observe more lifecycle events — subagent start/stop, tool execution, and turn metadata
Plugin sharing now exposes link metadata and discoverability controls — share/unshare individual plugins or entire workspaces, with source filtering and local share-path tracking
New codex remote-control command for headless app-server operation — orchestrate Codex sessions from CI, cron, or other agents without a TUI
App-server thread pagination with variable detail levels — list, peek, and replay long-running multi-agent threads efficiently
AWS Bedrock authentication now supports console-login credentials in addition to AWS profiles and access-key pairs
Multi-environment view_image — image references resolve across whichever environment the turn is running in
Live app-server threads now refresh from updated config without restart — change a model, profile, or hook and the next turn picks it up
Turn diffs remain accurate across nested apply_patch operations
Fixed remote-control retry handling under flaky network conditions
Persistent /goal workflows with app-server APIs, model tools, runtime continuation, and TUI controls for create / pause / resume / clear — goals round-trip across sessions and survive --resume
codex update command with configurable TUI keymaps and plan-mode nudges
Action-required terminal titles — iTerm2, WezTerm, and other compatible terminals reflect “needs input” / “running” in the title bar
Active-turn /statusline and /title editing without restarting the session
External agent session import including background imports and title handling — bring sessions from Claude Code or Aider into Codex with state preserved
MultiAgentV2 made explicit: thread caps, wait-time controls, root/subagent hints, v2-specific depth handling — no more silent unbounded fan-out
Documentation: Bundled OpenAI Docs skill updated for GPT-5.5 and gpt-image-2
Fixed stale interrupt hangs, persisted provider restoration, large remote response handling
September 2025: Skills system, configuration overhaul
October 2025: IDE extension launch, app-server API
November 2025: Cloud environments (Codex Cloud), Slack integration
December 2025: Linear integration, worktree support
January 2026: Enhanced review mode, automation improvements
February 2026: GPT-5.3-Codex, Codex Desktop App, steer mode stable, personal skills, parallel shell tools, memory plumbing
March 2026: Plugin system first-class, sub-agent addressing, thread search, userpromptsubmit hook, realtime sessions
April 2026: GPT-5.5 rolling out, in-app browser, computer use (macOS), thread automations, GitHub PR reviews, Amazon Bedrock provider, TUI reasoning controls, hooks stable, marketplace add
May 2026: codex doctor diagnostics, Vim editing in the composer, persistent goals on by default, codex remote-control foreground daemon, permission profiles, Python SDK auth
June 2026: Session archiving (/archive), clickable TUI links, F13-F24 keybindings, controller pairing, parallel standalone web search, Windows elevated sandbox setup