OpenAI Rolls Out a ChatGPT 'Library' for Enterprise and Opens 'Dreaming' Memory to Free Users

OpenAI spent the week of June 12 telling two very different audiences the same thing: ChatGPT is going to remember more, and it is going to organize what it knows. For enterprises, that promise arrived as a new "Library" — a shared shelf of reusable files and assets inside ChatGPT Enterprise, Edu, and Healthcare workspaces. For everyone else, it arrived as a quieter but arguably bigger shift: the company's "dreaming"-based memory system, until now reserved for paying Plus and Pro subscribers, is rolling out to Free users for the first time.

Taken together, the two launches sketch the shape of OpenAI's 2026 strategy — make the assistant stickier for individuals by deepening what it remembers, and make it indispensable for organizations by turning scattered uploads into governed, searchable institutional knowledge.

What the Library actually is

The Library, which began rolling out to Enterprise, Edu, and Healthcare workspaces around June 12, gives members "a dedicated place to find and reuse files they upload to or create in ChatGPT," according to OpenAI's release notes. In practice it functions like a workspace-level repository: a deck a colleague generated last quarter, a policy PDF someone uploaded, a spreadsheet the model produced — all collected in one place rather than buried inside individual chat threads where they effectively vanish after the conversation scrolls away.

Crucially, the Library is built for administrators, not just end users. Files saved to it follow the workspace's existing retention policies, and workspace owners can control whether ChatGPT automatically references Library files when it answers a prompt. That toggle matters: it is the difference between a passive file cabinet and an active knowledge base that quietly grounds the model's responses in an organization's own documents. The launch arrived alongside new admin controls in OpenAI's Cloud Console for managing Enterprise workspaces, reinforcing that this release is aimed squarely at IT and compliance teams rather than the casual user.

The same mid-June wave brought a batch of Codex upgrades for developers: improvements to Computer Use, a new Developer mode for deeper browser debugging, expanded access controls, clearer usage-limit guidance, and unread-chat navigation. The browser-debugging piece is the standout — Codex can now use controlled Chrome DevTools Protocol access to profile performance and inspect network traffic, console output, and runtime errors, much as a human engineer would. It is off by default and must be switched on in Codex settings.

"Dreaming" comes to Free users

The headline shift for the broader public is memory. ChatGPT's memory has, since June 4, been rebuilt on a system OpenAI calls "dreaming" — a background process that consolidates a user's entire conversation history into an evolving model of who they are, rather than the old, brittle list of manually saved facts. "Today we're beginning to roll out a more capable and scalable system for synthesizing memory, developed to tackle the staleness, correctness, and scalability challenges that we observe when memory is applied to the hundreds of millions of users and multi-year time horizons in ChatGPT," the company wrote.

The synthesized memories remain reviewable: a memory summary page lets users see the highlights of what ChatGPT knows, add or correct details, and set instructions for what topics it should raise and when. Plus and Pro users in the US got the upgraded architecture first, along with double the memory storage. Free users were the notable holdout — until now.

The reason for the wait was cost. "While dreaming-based memory has been available to Plus and Pro users for some time, we are only now able to offer Free users a version that meets our quality bar and is practical to serve at scale," OpenAI said. The unlock was an engineering one: "Recent improvements reduced the compute required to serve dreaming to Free users by approximately 5x, making it possible to begin rolling out dreaming to Free users over the coming weeks and to increase memory capacity for Plus and Pro users." Users who prefer the old behavior can switch back to legacy saved memories in settings.

Memory as a moat, knowledge as a hook

The strategic logic behind both launches is the same, viewed from opposite ends of the market. A chatbot that genuinely remembers your projects, your writing style, and the constraints you mentioned three months ago is far harder to abandon than one that greets every session as a stranger. Extending that to Free users — the top of OpenAI's funnel, where switching costs are lowest and competition from Google's Gemini and Anthropic's Claude is fiercest — is a deliberate bet that accumulated personal context will keep users from wandering. The 5x compute reduction is what made giving away that stickiness economically tolerable; serving memory to a free tier of hundreds of millions only works if each user's "dream" is cheap to compute.

The Library plays the same game in the enterprise. Its real value is not file storage but the gradual capture of an organization's institutional knowledge inside OpenAI's product, governed by the customer's own retention and access rules. Once a company's decks, contracts, and reference docs live in the Library and the model is grounding answers in them, ChatGPT stops being a generic assistant and becomes a system of record — exactly the territory Microsoft 365 Copilot, Glean, and Notion AI are fighting over. The admin controls and Cloud Console additions are the price of admission to that fight: enterprises will not feed proprietary documents into a tool they cannot audit and govern.

There are open questions. Background memory consolidation raises obvious privacy and accuracy concerns — a model that infers a profile from years of chats can also infer wrong things, and the burden of correcting them falls on users who may never check the summary page. For organizations, auto-referencing Library files is powerful until the model surfaces a stale policy or a confidential document to the wrong team.

What to watch next

The near-term tell is the rollout pace: OpenAI framed Free-tier dreaming as landing "over the coming weeks," so watch whether it ships globally or stays US-first like the Plus and Pro launch. On the enterprise side, the question is whether the Library evolves from a passive repository into a true retrieval layer with connectors to Drive, SharePoint, and internal wikis — the feature set that would let it directly challenge Glean and Copilot. And with the Enterprise and API "memory_context" parameter reportedly letting developers tap dreaming-style long-term profiles, the next competitive front may not be ChatGPT at all, but the third-party apps built to remember you on OpenAI's behalf.