ChatGPT Can "Dream" Now — Here's What Its New Memory Actually Changes

Dreaming is a background process that lets ChatGPT automatically curate and refresh its memory by referencing your chat history, rolling out from June 4, 2026 to Plus and Pro users and reaching free accounts over the following weeks. It fixes stale and incorrect memories inside ChatGPT, but it doesn't follow you to Claude, Gemini, or your coding tools — a shared memory layer does.

Dreaming (OpenAI calls the architecture Dreaming V3) tackles three problems the older saved-memories approach couldn't: staleness, correctness, and scale across hundreds of millions of users. Instead of only storing facts when you ask, ChatGPT now synthesizes its memory state in the background from past conversations. It carries preferences and constraints forward more reliably, and it updates entries as time passes — revising "You're going to Singapore in July" to "You went to Singapore in July 2026" once the trip is over. A memory summary page lets you see what it knows, edit it, and steer which topics it raises. By cutting compute roughly 5x, OpenAI brought memory to free accounts and doubled storage for Plus and Pro.

That's genuine progress. But three limits remain by design. The memory lives inside ChatGPT — open Claude, Gemini, Cursor, or any coding agent and none of it follows. The data is held on OpenAI's platform, not owned and exportable by you with an audit trail. And it's auto-synthesized: you review a summary, but you don't keep a verbatim, version-controlled source of truth. Those are exactly the gaps a cross-model memory layer fills.

You'll need:

• A free MemoryLake account
• A supported browser — Chrome, Firefox, or Edge
• The context you want available everywhere — preferences, rules, or files (PDF, Word, Excel, PowerPoint, text/Markdown, or images)

Follow these steps to keep one memory that works in ChatGPT and beyond:

1. Create your account

Create a free MemoryLake account.

1. Install the browser extension

Add the MemoryLake browser extension to Chrome, Firefox, or Edge.

1. Open ChatGPT

Go to chat.openai.com. The extension activates automatically, and a MemoryLake icon appears next to the message box.

1. Save what matters

Let auto-capture store context as you chat, or click Save on any message you want kept. Because the same memory lives in a MemoryLake Project, you can also point Claude, Cursor, or any MCP tool at it, so your context is shared rather than trapped in one app. For the full walkthrough, see the ChatGPT setup guide, and the MCP setup guide to connect other tools.

• Keep Dreaming on for in-ChatGPT convenience, and use MemoryLake for the context you need everywhere else, too.
• Store durable rules as named Memory entries so they recall precisely instead of from an auto-summary.
• Keep separate Projects for separate contexts (work, a side project) so each AI pulls only what's relevant.
• Point Cursor or Claude Desktop at the same Project to carry identical context into your editor.

• The MemoryLake icon doesn't appear: refresh chat.openai.com and confirm the extension is enabled for the site in your browser settings.
• Nothing is being saved: check that auto-capture is on, or save messages manually with the Save button.
• A saved memory is wrong: open your MemoryLake dashboard to edit or delete it — changes apply to the next chat.

Let Dreaming sharpen ChatGPT, and let MemoryLake carry the same memory into every other AI you use.