Codex Today

last sync: August 17, 2026 · 10:39 PM ET• sources: official + named staff• mode: dense

$ cat ./today.json

22 topics / direct sources

01ProductComputer History is opt-in on Mac desktop

Codex and ChatGPT can use a searchable activity timeline after you turn the feature on; it is not a default or global rollout.

deeper context / thread read

[OpenAI Developers]@OpenAIDevs says Computer History is opt-in so ChatGPT can pick up recent work, notice patterns, and suggest skills or scheduled tasks.
[OpenAI Developers]A follow-up limits the first rollout to the ChatGPT desktop app on Mac for Pro, Business, and Enterprise, with EEA, UK, and Switzerland access later; Business and Enterprise need an admin enablement first.
[Codex documentation]Official docs: off by default, requires Memories, unavailable with API keys or Amazon Bedrock, excludes the EEA, Switzerland, and the UK, and does not put screenshots or audio into history.
[ChatGPT & Codex changelog]The August 13 changelog repeats the same plan, admin, and regional constraints.

[OpenAI Developers] original source ↗

02ProductChatGPT desktop app for Linux is in preview

Supported Ubuntu, Debian, and Fedora desktops can run Chat, Work, and Codex locally; Computer Use is still missing on Linux.

deeper context / thread read

[OpenAI]@OpenAI announced a Linux desktop preview for ChatGPT, ChatGPT Work, and Codex on supported systems.
[OpenAI Developers]@OpenAIDevs lists Ubuntu 24.04/26.04 LTS, Debian 13, and Fedora 43/44, with .deb or .rpm packages for x64 and ARM64.
[Codex documentation]Linux docs add install, update, and Wayland notes, and say Computer Use is on macOS and Windows but not yet in this preview.

[OpenAI] original source ↗

03WorkflowDesktop and CLI can import other agents

Claude Code, Claude Cowork, and Cursor setups can be imported and optionally kept in sync; this is a migration path, not a merge of products.

deeper context / thread read

[OpenAI Developers]@OpenAIDevs says the ChatGPT desktop app can import projects, chats, skills, and plugins, show import history, and opt into automatic updates.
[ChatGPT & Codex changelog]The August 11 changelog names Claude Code, Claude Cowork, and Cursor, plus CLI `/import` for supported setup and recent chats.
[What's new]The weekly digest points to the import docs and the August 11 desktop and CLI notes.

[OpenAI Developers] original source ↗

04ProductDaybreak splits Blue and Red access

GPT-5.6-Cyber sits behind separately approved Daybreak Red; Blue is general defensive work and does not grant Red.

deeper context / thread read

[OpenAI]@OpenAI expanded Daybreak and introduced GPT-5.6-Cyber for advanced, authorized cybersecurity work.
[OpenAI]Daybreak Red is described as purpose-trained access, including GPT-5.6-Cyber, for authorized vulnerability research and testing.
[Greg Brockman]Greg Brockman pointed to the same Daybreak expansion and GPT-5.6-Cyber release.
[ChatGPT & Codex changelog]Codex changelog: Trusted Access for Cyber is required; approval is scoped to identity, workspace or API org/project, model, and surface; Blue does not include Red.

[OpenAI] original source ↗

05ProductCodex Security Review is a research preview

Eligible paid workspaces can request deeper GitHub PR security reviews; Plus is excluded and limits can apply.

deeper context / thread read

[OpenAI Developers]@OpenAIDevs launched Codex Security Review in research preview, using repository context to post findings on GitHub PRs.
[Greg Brockman]Greg Brockman said Codex can review every GitHub pull request and leave findings inline.
[What's new]What's new: enable automatic reviews on open or new commits, or mention `@codex security review`; Enterprise, Business, Edu, and Pro only.

[OpenAI Developers] original source ↗

06BuildersCodex Security CLI is public and early

The open-source CLI can scan repos and CI, but OpenAI still calls it an early release and scans still need Security access.

deeper context / thread read

[OpenAI]@OpenAI said Hacker News found the open-source Codex Security CLI first; it can scan repos, track findings, verify fixes, and add CI checks.
[OpenAI]Install path given by OpenAI: `npm install @openai/codex-security` or `npx @openai/codex-security@latest --help`.
[What's new]Docs say the public CLI/SDK reached 0.1.5 with separate versioning from the plugin; running scans still requires Codex Security access.

[OpenAI] original source ↗

07BuildersCodex CLI 0.147.0 adds plugins and auto-review

Stable CLI 0.147.0 is the documented feature cut; GitHub is already shipping 0.148.0 alphas without matching notes.

deeper context / thread read

[ChatGPT & Codex changelog]0.147.0 adds portable Agent Plugins, ordered chat sections, `--approve-for-me`, Claude/Cursor import without duplicate chats, MCP 2026-07-28, and Bedrock cached search plus remote compaction.
[OpenAI Codex on GitHub]GitHub tag rust-v0.147.0 is the referenced stable release.
[OpenAI Codex on GitHub]Pre-release rust-v0.148.0-alpha.21 was published August 17 with no public feature notes on the release page.

[ChatGPT & Codex changelog] original source ↗

08Signal, not release1M-token Sol can now use ChatGPT auth

Tibo documented a config override and later said ChatGPT accounts can use it; defaults stay smaller because OpenAI tuned them for cost and quality.

deeper context / thread read

[Tibo]Tibo posted `model_context_window = 1000000` and `model_auto_compact_token_limit = 900000` for gpt-5.6-sol, citing a documented 1,050,000-token window and warning the default is intentional.
[Tibo]A follow-up says the same override now works for ChatGPT-account usage, not only API keys.
[SamG]Replies asked for UI presets such as Balanced, Large Codebase, and Long-Running Investigation instead of raw TOML.

[Tibo] original source ↗

09CommunityTibo asked what Codex should already have shipped

The prompt is a request for obvious missing work, not a roadmap. Recurring replies so far center on usage, desktop smoothness, and stale subagent status.

deeper context / thread read

[Tibo]Tibo asked for things that are 100% within reach for Codex, the API, or models that OpenAI has not done yet.
[MidasAvocado]One highly engaged reply says the subagents bar can show agents as working when they are not until the user opens the subagent.
[Blake]Another recurring complaint is unexpected usage burn despite a high reported cache rate.
[xyzbk]A third asks for ChatGPT desktop performance work, citing stuttering on a high-end PC.

[Tibo] original source ↗

10Signal, not releaseStaff say auto-compaction is now the default path

Codex leadership is telling people to stop managing context windows by hand; that is a usage claim, not a new product ship.

deeper context / thread read

[Andrew Ambrosino]Andrew Ambrosino said he has not thought about context windows in months and called Codex auto-compaction good enough to turn the usage donut off.
[Romain Huet]Romain Huet separately called Codex auto-compaction “pretty magical.”
[Tibo]Tibo still documents a larger window as an override, not the recommended default.

[Andrew Ambrosino] original source ↗

11Signal, not releaseCodex now has two subagent tiers

pvncher says Sol and Terra can collaborate and recurse; Luna and older models are leaf delegates only.

deeper context / thread read

[eric provencher]pvncher distinguishes full collaborative agents (Sol and Terra) from leaf Luna or older-model delegates.
[eric provencher]He added that Luna agents can fork but lack the same cross-agent coordination and are not peer agents.
[eric provencher]A later post says prompting custom multi-agent behavior matters more now that Luna can take well-scoped bounded work.

[eric provencher] original source ↗

12RemoteVoice can steer Codex from desktop and iOS

GPT-Live Voice is live on desktop globally for paid plans; iOS Remote voice exists, and Android voice in Codex was still “coming soon” in July.

deeper context / thread read

[OpenAI]@OpenAI said ChatGPT Voice in the desktop app can direct Work or Codex agents via GPT-Live, rolling out globally on macOS and Windows to Plus, Pro, Business, Edu, and Enterprise.
[OpenAI]A same-day follow-up said Voice also works in Codex from the iOS app with paired remote access, with Android support coming soon.
[What's new]Docs: Voice can start or steer other threads; on macOS, “Take a look at this” can share an appshot when Screen context is on.

[OpenAI] original source ↗

13RemoteRemote on iOS remains the mobile control surface

Staff are amplifying start-from-iOS Remote, not announcing a new mobile app. Community asks include widgets, Files, and system apps.

deeper context / thread read

[Thomas Ricouard]Dimillian highlighted Codex Remote after Federico Viticci described start-from-iOS, near-desktop parity, and voice dispatch across threads.
[Atty Eleti]Atty Eleti called Codex Remote “a glimpse of the future.”
[Federico Viticci]Viticci’s concrete asks were Home Screen widgets for Remote and voice, Reminders/Calendar integrations, and Files-app folder mounts.
[ChatGPT & Codex changelog]iOS 1.2026.209 changelog focuses on task reconnection, prompt preservation, voice background settings, and approval recovery after reconnect.

[Thomas Ricouard] original source ↗

14WorkflowUnified Exec is default Codex on Windows

Derrick Choi says Windows now matches the macOS/Linux command model for long-lived processes; treat this as a staff confirmation, not a dated changelog cut.

deeper context / thread read

[Derrick Choi]Choi said Unified Exec is now the default for Codex on Windows, so Codex can start a dev server, keep it running, watch output, send later input, and run test watchers.
[Derrick Choi]He separately asked APAC users, in Chinese and English, what they most want improved in Codex.
[Codex documentation]Windows desktop docs remain the official surface for the ChatGPT/Codex Windows app.

[Derrick Choi] original source ↗

15ProductCodex now lives inside the ChatGPT desktop app

The July 9 merge is still the current desktop shape: Chat, Work, and Codex share one app; old Codex app users update in place.

deeper context / thread read

[OpenAI]@OpenAI introduced ChatGPT Work as an agent in ChatGPT powered by Codex and GPT-5.6.
[OpenAI]Availability: Work on web/mobile started with Pro, Enterprise, and Edu, then Plus and Business; desktop Chat, Work, and Codex are on every plan, including Free, on Windows and Mac. Codex app users update into the ChatGPT desktop app.
[What's new]What's new: dedicated Codex view, inline diffs, PR review, faster Computer Use on GPT-5.6, multi-repo projects, and optional Codex default view/icon.

[OpenAI] original source ↗

16WorkflowAuto-review moved to GPT-5.6 Luna

OpenAI expects Auto-review to cost about 10x less after the Luna switch; it reviews eligible actions, it does not widen sandbox rights.

deeper context / thread read

[OpenAI]@OpenAI said Auto-review in the ChatGPT app and Codex CLI moved from GPT-5.4 to GPT-5.6 Luna, with expected cost about 10x lower.
[What's new]CLI 0.147.0 adds `--approve-for-me` for automatically reviewed approvals without expanding filesystem or network permissions.
[Codex documentation]Auto-review docs remain the official constraint page for the feature.

[OpenAI] original source ↗

17WorkflowGPT-5.4 leaves ChatGPT-signed Codex on August 31

ChatGPT-authenticated Codex must move defaults to 5.6 Terra/Luna; API-key Codex is not affected.

deeper context / thread read

[What's new]On August 31, GPT-5.4 and GPT-5.4 mini retire from Codex for ChatGPT sign-in. Replace `gpt-5.4` with `gpt-5.6-terra` and `gpt-5.4-mini` with `gpt-5.6-luna` in defaults, agents, and scheduled tasks.
[What's new]Recommended local/web models remain Sol, Terra, and Luna; Power defaults to Sol with medium reasoning.
[Codex documentation]Model catalog and workspace availability pages are the admin check before the cutoff.

[What's new] original source ↗

18ProductSites and role plugins still define non-dev Codex

June 2 remains the source for role plugins, in-place annotations, and Sites preview for Business and Enterprise.

deeper context / thread read

[OpenAI News]OpenAI said more than 5 million people use Codex weekly and launched six role plugins covering 62 apps and 110 skills.
[OpenAI News]Sites preview lets Business and Enterprise teams host interactive pages by URL; Enterprise admins enable Sites in admin settings.
[OpenAI News]Annotations now apply beyond code to documents, spreadsheets, slides, and sites by pointing at a selected region.

[OpenAI News] original source ↗

19Signal, not releaseOpenAI published Codex usage as economic evidence

The June 25 paper is research, not a product launch; task-hour figures are model-estimated from a sample.

deeper context / thread read

[OpenAI News]By May 2026, OpenAI estimates 70.2% of sampled individual users made at least one Codex request that would take a person more than an hour.
[OpenAI News]OpenAI says Codex is now the primary internal AI tool across departments, including Legal and Recruiting.
[OpenAI Economic Research]The underlying paper is posted as a PDF.

[OpenAI News] original source ↗

20BuildersAgent Plugins lists Codex among launch clients

OpenAI Dev named Codex as a compatible client of agent-plugins.org; that is interoperability, not a Codex feature release.

deeper context / thread read

[OpenAI Developers]@OpenAIDevs listed Codex, ChatGPT, Cursor, GitHub Copilot, Kiro, and VS Code as compatible agentic clients at launch.
[Agent Plugins]The directory they pointed to is agent-plugins.org.
[ChatGPT & Codex changelog]CLI 0.147.0 is the Codex-side cut that searches local, personal, workspace, and remote plugin catalogs.

[OpenAI Developers] original source ↗

21CommunityBuild Week and education plugins stay on the calendar

These are still-relevant programs, not this week’s ship. Ambassador applications remain paused per meetup docs.

deeper context / thread read

[OpenAI Developers]@OpenAIDevs posted that Build Week participants shipped real projects with Codex.
[What's new]What's new documents College Student, College Educator, and K–12 Educator plugins for ChatGPT Edu and Teachers deployments.
[Codex Meetups]Meetup listings continue through late August; ambassador applications are paused.

[OpenAI Developers] original source ↗

22RemoteGPT-5.6 on Bedrock is a local Codex path

Sol, Terra, and Luna can run through Amazon Bedrock in local Work and Codex surfaces; this is not cloud Codex hosting.

deeper context / thread read

[What's new]GPT-5.6 Sol, Terra, and Luna reached GA on Amazon Bedrock for local ChatGPT Work and Codex, including desktop, CLI, IDE, and SDK, via a Bedrock API key or the AWS SDK chain.
[Codex documentation]Computer History explicitly does not work with Amazon Bedrock.
[Codex documentation]Bedrock is documented as the enterprise provider page for Codex.

[What's new] original source ↗