ChatGPT Enterprise workspaces can now share Codex plugins by default

OpenAI has quietly flipped an important default for ChatGPT Enterprise customers using Codex. Plugin sharing, previously something you had to request activation for through your OpenAI account contact, is now switched on by default for eligible Enterprise workspaces. If your team uses Codex, this changes how you can distribute internal tools and workflows starting today.

What Changed, Exactly

Until now, Enterprise users who wanted to share locally built plugins with colleagues had to ask OpenAI to enable the feature. ChatGPT Edu workspaces already had it on by default, but Enterprise did not. That gap is now closed.

With sharing enabled by default, any Codex user in an eligible Enterprise workspace can publish a local plugin to a shared workspace directory. Teammates can then find it under “Shared with you” in the Codex plugin directory and install it themselves. No IT ticket, no waiting on OpenAI.

The sharing stays inside your organisation. Shared plugins are not published to a public directory, and anyone outside your workspace cannot access them.

What Codex Plugins Actually Are

If you have not spent time with Codex plugins yet, a quick primer helps here.

A plugin is a self-contained bundle that packages together skills (reusable task definitions), app integrations, MCP server configurations, and lifecycle hooks into a single installable unit. The idea is that instead of every team member manually setting up the same Codex environment, one person builds it once, packages it as a plugin, and shares it. Everyone else installs it in a few clicks.

Each plugin has a manifest at .codex-plugin/plugin.json and can include a skills/ directory, an .app.json file pointing at connected apps or connectors, an .mcp.json file for MCP server config, and assets for how the plugin appears in the directory.

The plugin directory itself groups content into three sections:

Curated by OpenAI — highlighted plugins available to all Codex users
Shared with you — plugins shared by workspace members
Created by you — plugins you have built or added yourself

To share a plugin you have built, you open the Codex app, go to Plugins, find the plugin under “Created by you,” open its details page, and select “Share.” From there you can add specific workspace members, add workspace groups, or copy a share link. You choose who gets access.

Groups are the right tool when an entire team or role should share the same plugin. If you want broader repo or CLI distribution, a marketplace setup makes more sense. Workspace sharing is specifically for getting teammates to install a plugin through the Codex app.

Recipients find shared plugins under “Shared with you” in the directory and install from there.

What This Means for Enterprise Admins

The default-on change shifts the administrative posture from opt-in to opt-out. If your organisation needs to restrict plugin sharing, you now need to actively disable it rather than simply never enabling it.

The control lives in requirements.toml. Admins can set plugin_sharing = false in cloud-managed requirements to turn off workspace sharing for locally built plugins. This can be deployed via MDM (using com.openai.codex:requirements_toml_base64 on macOS) or through the Codex Policies page, which lets admins push enforced configuration to specific groups without distributing device-level files first.

Codex applies requirements layers in a defined order: cloud-managed requirements take priority, then MDM-managed preferences, then system-level requirements.toml files. So a cloud-managed plugin_sharing = false will override anything set at the device level.

Beyond the sharing toggle, admins and owners have further controls available. You can disable specific plugins by disabling the corresponding app in Workspace settings under Apps. Enterprise and Edu admins can also use RBAC to control which users get access to a given app or plugin. One current limitation: app controls apply across all surfaces at once, including ChatGPT web, mobile, and Codex. You cannot enable a plugin in Codex while disabling it in the ChatGPT mobile app.

The Broader Picture

This change fits into a deliberate effort to make Codex a practical team tool rather than something only developers configure individually. OpenAI has noted that roughly 20% of Codex’s weekly users now come from non-engineering roles, including analysts, marketers, and operations teams, and that group is growing faster than the traditional developer base.

Plugins are a direct response to that shift. When someone on the finance team builds a Codex workflow that queries internal data and formats a report, packaging it as a plugin means everyone with the same need can use the same setup. Before this change, sharing that plugin across an Enterprise workspace required a manual handshake with OpenAI. Now it just works.

For teams that have been building internal Codex tools and emailing colleagues instructions on how to replicate their setup, this is a meaningful quality-of-life improvement. The infrastructure was already there. The default finally matches how people actually want to work.

Updates to this story

3 July 2026: ChatGPT Business adds centralised plugin management in Workspace settings with governance, filters, and install controls

The June 2025 article covered plugin sharing being switched on by default, giving teams a way to discover each other’s locally built plugins. This update goes a layer deeper: how admins govern all plugins across the workspace from a single control surface.

OpenAI has added a dedicated Workspace Settings → Plugins page for Business, Enterprise, and Edu workspaces. Admins and owners can now search and filter all plugins by status, installation policy, role, category, and catalog, and set per-plugin install policies from one place rather than managing plugins user by user. For Enterprise and Edu plans, workspace owners can also target install policies to specific roles via Workspace Settings → Permissions & Roles.

To disable a plugin, admins disable the corresponding app from Workspace Settings → Apps, or use the Manage Actions menu to restrict which actions the plugin can perform. These controls apply across all surfaces, including ChatGPT web, mobile, Atlas, and Codex, so it is not currently possible to enable an app and its plugin in one surface but not another.

A plugin does not grant new data access on its own. Users can only use app-backed capabilities when the connected app is available to them and they already have access to the underlying source.

Admins rolling out Codex at scale should audit existing app settings and RBAC access against their workspace needs. Full details are in the ChatGPT Business release notes and the Plugins in Codex admin reference.

23 June 2026: Codex update: rollout token budgets, organised plugin discovery, and /usage credit redemption

The June 23 Codex release builds directly on the plugin-sharing rollout covered below and adds two further changes worth noting.

On plugins, the /plugins screen now organises remote plugins into three labelled sections: OpenAI Curated, Workspace, and Shared with me. The “Shared with me” section makes workspace-shared plugins a first-class destination in the directory, replacing the flat list that existed when default plugin sharing launched. Discovery no longer requires browsing an unsorted catalogue.

The more operationally significant addition is configurable rollout token budgets. A single token ledger now tracks usage across all agent threads under one rollout, delivers in-thread reminders as the budget depletes, and aborts turns when the limit is exhausted. Fields including limit_tokens, reminder_interval_tokens, sampling_token_weight, and prefill_token_weight are available in config.toml. This is a direct response to longstanding user reports of faster-than-expected credit consumption under token-based pricing (changelog).

The /usage command can now display daily, weekly, and cumulative token activity and redeem earned usage-limit reset credits with confirmation and retry states, removing the need to visit a separate settings page.

The core story below remains accurate; these additions extend rather than supersede it.

23 June 2026: ChatGPT Enterprise adds Slack connector actions — join channels, create reminders, upload files from ChatGPT

ChatGPT Enterprise and Edu workspaces can now enable write actions through the Slack connector, expanding well beyond the search capability that existed at the time this article was originally published. Supported actions now include joining a channel, creating a reminder, uploading a file, and updating a user’s Slack profile. Details are in the ChatGPT Enterprise & Edu release notes.

The connector remains off by default. Admins enable it through Workspace connector settings and use Action control to govern which specific actions are available. Write actions are set to “Always ask” by default during agent runs. Some new actions require additional Slack OAuth scopes, meaning a re-authorization step from a Slack workspace or Enterprise Grid admin may be needed. Existing read-only search continues to work without those additional scopes. The Slack app operates across Chat, Deep Research, and Agent Mode.

This update also lands in a more competitive context than the original post addressed. Anthropic released Claude Tag on the same day, a Slack-native tool built around an ambient, always-on model that proactively monitors conversations and follows up on tasks. The two products take different approaches: OpenAI extends a governed connector with discrete write actions, while Anthropic positions Claude Tag as a persistent AI teammate inside Slack. The ChatGPT connector for Slack help article covers full setup requirements.

18 June 2026: OpenAI Launches Credit Usage Analytics and Granular Spend Controls for ChatGPT Enterprise

OpenAI has added credit usage analytics and tiered spend controls to the Global Admin Console for ChatGPT Enterprise, available immediately to all enterprise customers. The announcement is at openai.com/index/chatgpt-enterprise-spend-controls.

The new dashboard surfaces ChatGPT and Codex credit consumption broken down by user, product, and model in a single view, with trend tracking over time and access to the same data via a unified Cost API for internal BI tools. Admins can now set hard monthly credit limits at three levels: workspace-wide defaults, group-level caps, and individual user overrides. Individual contributors can view their own usage against their allocated budget and request additional credits without requiring a department-wide limit increase.

This directly extends the context of the plugin-sharing update from earlier this month. With Codex plugin sharing now on by default, enterprise teams are adopting Codex faster, and the new spend controls address the predictability problem that creates. Zipline, which expanded Codex company-wide in 2026, explicitly asked OpenAI for these tools to identify low-adoption users and keep spend predictable as usage scaled.

The practical shift for enterprise buyers is that AI costs can now be metered and allocated internally, making chargeback and departmental budgeting viable and reducing the risk that a large bill triggers a blanket usage freeze.