ChatGPT Space: OpenAI’s Shared Workspace for Pages, Teams, and AI Agents
ChatGPT Space: OpenAI’s Shared Workspace for Pages, Teams, and AI Agents
ChatGPT has been good at producing work for a while. The awkward part has been what happens next.
A useful answer gets copied into a document. Research disappears into one person's history. A brief goes stale. An agent finishes, but its result has no obvious home, owner, or update path.
ChatGPT Space is OpenAI's attempt to fix that middle layer.
Space is a shared home for Pages, files, people, ChatGPT, Codex, and dots. A team can co-edit a page, comment, mention an agent, pull context from connected tools, and ask an automation to keep the artifact current. It replaces Library for eligible users. That makes it look like a document and collaboration product, but the bigger idea is shared state for human and agent work.
That is where the hard questions begin. If AI edits a launch plan overnight, who owns the change? Which source justified it? Can the team restore the last approved version? Does revoking a Drive file remove its summary from the page?
This article covers what shipped at OpenAI DevDay 2026, what is still coming, and how teams should design ownership, permissions, provenance, and rollback before treating a living page as a source of truth.
What is available on September 29, 2026?
Fact layer, checked September 29, 2026: OpenAI announced Space, Pages, teams, Team Tasks, collaborative slides, and workplace mentions together, but they do not share one launch boundary.
| Capability | Plans | Status at launch |
|---|---|---|
| ChatGPT Space and Pages | Pro, Business, Enterprise | Create, edit, and collaborate on web and desktop |
| Space on mobile | Pro, Business, Enterprise | Find, read, and share; creation, editing, and collaboration are coming later |
| Collaborative slides | Pro, Business, Enterprise | Coming in the following weeks, with PowerPoint and Google Slides export planned |
| Collaborative spreadsheets | Pro, Business, Enterprise | Listed as coming soon, not a launch-day capability |
| Teams and Team Tasks | Business, Enterprise | Available for shared groups and recurring or event-triggered work |
| @ChatGPT in Slack and Microsoft Teams | Business, Enterprise | Available, subject to workspace, app, and user permissions |
| Meetings plugin | Pro, Business | Beta on the macOS desktop app; Enterprise is coming soon |
The primary sources are OpenAI's DevDay recap and the current ChatGPT Space product page. OpenAI says rollouts and workspace availability remain subject to administrator settings.
Space, Pages, Work, and dots: four different jobs
Product names get confusing because all four surfaces can touch the same project. The products overlap, but each owns a different part of the workflow:
ChatGPT Work = executes a delegated task
dot = holds an ongoing responsibility
Page = the editable artifact humans and agents work on
Space = the shared context and organization around those artifacts
That distinction matters for RohitAI readers because Space is a sequel, not a repeat, of our earlier ChatGPT Work analysis.
Work answered: Can ChatGPT gather context, use tools, and finish a substantial task?
Space answers: Where does that work live, who can change it, and how does a team continue from it?
An OpenAI dot can monitor customer feedback. Work can analyze it and draft a launch plan. Codex can turn the plan into engineering changes. A Page can hold the brief, evidence, decisions, and comments. Space brings those actors back to the same context.
Space becomes useful when contribution and control travel together. A shared artifact without source history, ownership, and recovery is only a faster way to create confusion.
Pages are living documents, which means they behave like software
OpenAI describes Pages as documents built for work with agents. They can contain text, images, charts, code, and interactive tools. Teammates can edit in real time, leave comments, and mention ChatGPT or a dot to make changes. Instructions can tell ChatGPT what to update, which sources to use, and when to check.
It looks like a collaborative document. Operationally, it is closer to a lightweight application with a document interface.
A page can have:
- inputs from connected systems;
- instructions that function like business logic;
- scheduled or event-driven updates;
- multiple human and agent writers;
- permissions that determine who sees and changes the result;
- downstream readers who treat it as current.
That means a page needs some software disciplines. Instructions should be testable. Sources should be named. Automations should have owners. Changes should be reviewable. A bad update should be recoverable.
OpenAI's launch materials do not establish that every requirement for version comparison, approval, retention, or rollback is solved. Verify those controls in your workspace before making a Page authoritative for a policy, financial plan, release checklist, or incident runbook.
The permission boundary is not where most people think it is
The Space FAQ documents an important distinction.
An uploaded file follows the Page's permissions. A linked file keeps its own permissions; linking it does not grant collaborators access. But if ChatGPT reads an authorized source and copies or summarizes its content onto the Page, that new text is visible to everyone who can access the Page.
| Action | What collaborators receive | Risk to review |
|---|---|---|
| Share a Page | The Page and its visible contents, at view or edit level | Generated text may include details drawn from private context |
| Upload a file into a Page | The file follows the Page's permissions | A previously private file may now have a broader audience |
| Link an external file | The link; underlying access remains controlled by the source system | Users may mistake a link for shared access |
| Copy or summarize external content | The copied material becomes Page content | Source permissions no longer protect the derived text on the Page |
| Use personal memory while editing | Private memory is not directly shared | Anything written from that memory is visible to collaborators |
The key governance consequence is that AI turns access into content movement.
Traditional permissions ask whether Alice can open a file. An agent system must also ask whether Alice's ChatGPT can transform that file into a summary and place the summary somewhere Bob can read. The source file can remain perfectly permissioned while its substance crosses a boundary through generated text.
The safe control is not “never summarize.” It is to classify destination Pages, restrict sensitive source combinations, and require review before moving content into a broader audience.
Sharing does not expose private memory—but output can
OpenAI says sharing a Page does not grant collaborators access to your private chats or personal memory. That is the right boundary.
The warning immediately after it matters more: anything you or ChatGPT write on the Page is visible to people with access, including personal details drawn from memory.
This is a subtle but important distinction:
private context remains private as a data store
generated content can carry facts out of that context
Before sharing, review the artifact itself. Do not rely on the privacy of the originating memory or chat to protect text that has already been written into a shared Page.
For individual accounts, training and memory settings apply. OpenAI says it does not train on Business or Enterprise data by default. Those are data-use policies, not replacements for page-level sharing discipline.
Our previous analysis of OpenAI Computer History and Codex desktop memory made the same broader point: once context can follow work, provenance becomes more important than the convenience of remembering it.
Ownership: every living page needs a human editor
Space lets multiple people work on one Page, and OpenAI says each person uses their own ChatGPT when asking questions or making changes. This preserves personal interaction while editing a shared artifact. It also creates a coordination problem: a Page may contain human text, one person's ChatGPT output, another person's dot update, a Team Task, and imported tool data.
Without explicit ownership, “the AI updated it” becomes an excuse for nobody being accountable.
Every operational Page should display:
human owner
approved audience
authoritative sources
last human review
next review deadline
active automations
material changes since approval
The ownership rule is simple: agent identity is not artifact ownership. A dot can make the edit, but it should not decide whether the resulting policy, budget, roadmap, or external statement is approved.
This is where OpenAI's wider enterprise agent platform matters. We explored the identity and coordination layer in OpenAI Presence. Space makes that abstract issue visible inside an everyday document.
Provenance is more valuable than polished prose
An AI can rewrite a rough plan beautifully. That is not the same as making it trustworthy.
For a living Page, the important questions are:
- Which source supports this claim?
- When was the source last checked?
- Was the value copied, calculated, or inferred?
- Which person or agent changed it?
- Did an approved reviewer accept the change?
- What was the previous value?
OpenAI encourages Page instructions that specify sources and timing. Teams should go one step further and make evidence part of the artifact. Require links beside factual claims, a “last checked” field for volatile numbers, and a visible change note for agent-written updates.
The trust consequence is that shared context without provenance becomes shared confidence, not shared knowledge.
An agent can make a Page feel coherent even when its sources disagree. The more polished the document becomes, the easier it is for readers to forget which parts are facts, assumptions, and proposals.
Link the original system or document, record when it was checked, and preserve the exact value or statement the team relies on.
Name the author—human or agent—and show the assumptions that connect evidence to the conclusion.
Mark the decision owner, review deadline, alternatives, and the condition that moves the proposal into approved state.
Rollback is a product requirement, not an emergency feature
If a Page updates itself from Slack, email, a calendar, or a project tracker, it will eventually ingest something wrong, stale, premature, or adversarial. A source can be corrected. A project name can change. A Slack message can contain speculation. An automation can follow a technically valid but outdated instruction.
Before enabling unattended updates, verify what version history and restoration controls your plan and workspace expose. For high-impact Pages, also keep periodic approved snapshots or export the authoritative version to a system with established retention and rollback.
The safest pattern is:
agent proposes change
diff shows source and reason
human approves material change
approved state is checkpointed
readers see review freshness
rollback path is tested
Not every to-do list needs that machinery. A board briefing, pricing sheet, incident plan, policy, or release checklist often does.
Where Team Tasks fit
Business and Enterprise teams can use Team Tasks for recurring work such as weekly project updates. OpenAI says tasks can run on a schedule or in response to an event, such as a new email or Slack message, and teammates can refine the instructions together.
That makes Team Tasks the automation layer under Space.
It also raises a familiar software problem: shared editing of instructions can change production behavior. Teams should decide who can modify the automation, whether changes require review, and how the next run is tested.
RohitAI's analysis of OpenAI Skills across ChatGPT, Codex, and the API is useful here. Reusable instructions are portable workflow code. Space makes those workflows collaborative, but collaboration does not remove the need for versioning and permissions.
A practical operating model for Space
People and ChatGPT co-edit. No unattended updates. Best for briefs, research synthesis, and early planning.
Named sources and scheduled refreshes keep a dashboard or project page current. Material changes require human review.
dots, Team Tasks, Work, Codex, and connected tools update the same project state. Use only with explicit ownership, logs, budgets, and rollback.
Most teams should begin at Level 1. A polished demo may jump directly to Level 3, but reliability comes from earning automation one boundary at a time.
Cost and adoption: shared work needs a budget owner too
Space is available on Pro, Business, and Enterprise. OpenAI separately says Codex and ChatGPT Work tasks that a dot starts or manages continue to count toward those products' usual plan limits; verify current usage terms for other tools and automations.
For enterprise teams, the better unit is cost per accepted update or completed workflow. Our breakdown of OpenAI's enterprise Work and Codex rate card explains why an apparently simple artifact can sit on top of model tokens, tools, delegated workers, and reviews.
Space could reduce coordination cost dramatically. It can also hide compute behind a calm document interface. Put a budget owner beside the content owner.
Frequently asked questions
What is ChatGPT Space?
Space is the shared home for Pages and files created, uploaded, and shared in ChatGPT. It replaces Library for eligible Pro, Business, and Enterprise users and lets people work with ChatGPT, Codex, and dots around shared context.
What is a ChatGPT Page?
A Page is an editable, collaborative document that can combine text, images, charts, code, and interactive tools. Teammates can edit together, comment, and ask ChatGPT or a dot to make changes.
Is ChatGPT Space the same as ChatGPT Work?
No. Work is the execution surface for delegated tasks across tools, files, and browsers. Space is the shared context and artifact layer where a team organizes and continues that work.
Can I edit a Page on mobile?
Not at launch. As of September 29, mobile supports finding, reading, and sharing Pages. OpenAI says mobile creation, editing, and collaboration are coming later.
Does sharing a Page expose my private chats or memory?
No, not directly. But any content you or ChatGPT writes on the Page is visible to collaborators, including information the model drew from personal memory. Review generated content before sharing.
What happens to permissions when I add a file?
An uploaded file follows the Page's permissions. A link to an external file does not grant access to that source. However, content copied or summarized from the external file becomes visible as Page content.
Are collaborative slides and spreadsheets live now?
Pages are live. OpenAI says collaborative slides are coming in the following weeks and lists collaborative spreadsheets as coming soon. Do not plan a launch-day workflow around either until they appear in your account.
Can a dot update a Page automatically?
Yes. OpenAI says dots work across Space files and can keep Pages current as plans change. Team Tasks can also run scheduled or event-triggered work on Business and Enterprise plans. Set ownership and review rules before enabling unattended updates.
Conclusion: the source of truth needs a truth process
ChatGPT Space is more consequential than its document-shaped interface suggests.
It gives OpenAI a place where people, Pages, Work, Codex, dots, plugins, and Team Tasks can share project state. That could remove a large amount of copy-paste coordination and keep useful work from dying inside private chats.
But a shared artifact becomes valuable only when the team can trust how it changed.
The hard product is not real-time co-editing. It is the truth process around the page: who owns it, which sources count, where private context can flow, which agent made a change, what a reviewer approved, and how the team restores the last good state.
Start Space as a collaborative draft surface. Add named sources. Add one carefully scoped automation. Make agent contributions visible. Test revocation and rollback. Only then promote a living Page into the system your team relies on.
OpenAI has built the room where humans and agents can work together. Teams still have to decide who holds the keys.