OpenAI is moving ChatGPT closer to a persistent assistant by rolling out Memory to ChatGPT Plus users. The feature lets users tell ChatGPT what they want it to remember across chats, instead of treating every new conversation as a blank slate.
The update changes how ChatGPT can carry context from one exchange to the next. It also puts new weight on the controls users have over what is remembered, what is forgotten, and when memory should not be used at all.
Memory moves beyond a single chat
OpenAI has enabled the Memory feature for all ChatGPT Plus users, according to the company’s announcement via X. To use it, a user can start a new chat and tell ChatGPT what they want it to remember.
The core idea is simple: ChatGPT can retain context from previous conversations and bring it into future ones. Until this feature, a new chat began without prior knowledge from earlier conversations.
That matters because many ChatGPT tasks are repetitive. A user may regularly ask for writing help, coding support, planning, or business materials. If ChatGPT can remember preferences across sessions, it may need fewer reminders before producing a useful answer.
OpenAI says Memory can help ChatGPT learn preferences and style. The source examples include a preferred general tone, language, or format for blog posts; a preferred programming language and frameworks for coding; or preferred charts for monthly business meetings.
Users can manage what ChatGPT remembers
Memory is not designed as an invisible setting that users cannot influence. The source says users can control ChatGPT’s memory and instruct it to remember or forget certain details.
Those instructions can happen directly in chat or through the settings. The feature can also be turned on and off in settings.
For users who want to separate certain conversations from memory, OpenAI also provides temporary chats. These chats do not appear in history, do not use memory, and are not used to train OpenAI’s models.
That creates three practical modes of use:
- Use Memory: Tell ChatGPT what to remember so future chats can reflect those preferences or details.
- Use temporary chat: Keep a conversation separate from history, memory, and model training.
- Turn Memory off: Disable the feature completely in settings.
These controls are especially relevant when the same account is used for different kinds of work. The source notes that temporary chats should be particularly useful for teams that share an account but do not work on the same topics, because cross-chat memory could become a hindrance in that situation.
Availability is still uneven
The April 29, 2024 update says Memory is being rolled out for ChatGPT Plus accounts and has been enabled for all ChatGPT Plus users. But the feature is not available everywhere.
OpenAI said Memory is not "currently available" in Europe and Korea. The company did not give a reason in the source article.
The earlier February 13, 2024 article described Memory as a test available only to a few ChatGPT Free and Plus users. At that stage, it could be enabled in settings under "Personalization" and "Memory", with wider availability to be announced later.
OpenAI also plans to bring Memory to Team and Enterprise customers and GPTs in the future. For GPTs, the source says each GPT will have its own memory when the feature becomes more widely available.
What it means for teams and GPTs
For individual users, Memory is mainly about convenience. Instead of repeating a preferred format or work style, users can ask ChatGPT to remember it and apply it later.
For teams and companies, the stakes are different. Shared work often involves different projects, roles, and contexts. A memory that helps one workflow could confuse another if the same account is used across unrelated topics.
The source states that company and team user data, including memory, is excluded from OpenAI model training. It also says account owners can disable memory for their organization at any time.
That gives organizations a clearer administrative choice: allow memory where it improves repeated workflows, or turn it off where persistent context could interfere with how teams use ChatGPT.
GPTs add another layer. If each GPT has its own memory, a user’s interaction with one GPT would not necessarily behave like interaction with another. The source does not provide implementation details, but it does say Memory for GPTs will arrive as the feature becomes more widely available.
OpenAI has not explained the technical details
OpenAI has not released details of Memory’s technical implementation or privacy aspects. The source article says OpenAI does not specify how ChatGPT’s memory works.
The article describes one possible mechanism: it probably uses text mining to create a database of facts from previous conversations, which are automatically extracted and incorporated into new responses that match the user’s prompts. The source compares this to an extended, automated form of "Custom Instructions".
The broader direction is clear from the feature’s purpose. Memory is part of OpenAI’s effort to make ChatGPT work more like a personal assistant. The source also says the company is reportedly working on two agents for different use cases, and presents ChatGPT’s memory and networked GPTs as a preview of a fuller assistant package.
The idea itself is not new. In spring 2023, a developer presented MemoryGPT, a cross-chat memory based on the API. It used a combination of a vector database and a regular data store for high-level user data and goals.
For now, the practical takeaway is narrower. ChatGPT Plus users can begin using Memory by telling ChatGPT what to remember, while keeping control through settings, forget instructions, temporary chats, and the option to turn Memory off.