AI With Memory Like ChatGPT: Compare What Carries Over
Several AI assistants now save context across sessions, but they differ in how long that memory lasts, who controls it and whether it works across people and apps.
Updated 8 min read
Several AI assistants now save something between sessions, but what they save, how long it lasts and who can read it varies widely. ChatGPT memory stores facts the user shares about themselves and their preferences. Claude uses project instructions you write. Gemini can reference items in your Google Workspace. Each covers a different slice of continuity. For shared or long-running project work, a tool like fenbs keeps the record outside any one assistant so it survives a model change and is available to teammates.
Understanding the difference matters before you rely on any assistant's memory for something important. Personal memory in one app does not automatically reach your teammate's assistant, a different AI app you switch to, or a fresh session that starts without the right context loaded.
What memory means in an AI assistant
When an AI assistant appears to remember something, that information came from somewhere specific: the current chat, a saved preference, a project file, or instructions written into the system prompt. Those are different stores with different lifetimes and access controls. Knowing which one you are relying on helps you check whether a future session can actually use it.
True cross-session memory means the assistant can access information from a previous conversation without you pasting it in. Partial memory means the app stores some facts but not all context. Session-only memory means nothing persists once the conversation ends. Most consumer AI apps sit somewhere between partial and true cross-session memory, with controls that let users view and delete what is stored.
How ChatGPT saves context across sessions
ChatGPT memory stores facts the user declares or that the model decides are worth keeping: name, job, preferences, recurring tasks. You can view and edit these notes in Settings under Personalization. Memory is per-user and stays in your account; it does not transfer to another person's account or to the ChatGPT API by default.
ChatGPT Projects add a second layer: you can write project instructions and upload files that stay active for every conversation inside that project. For information about how ChatGPT projects handle memory, see the ChatGPT project memory guide. Note that Temporary Chat mode disables memory for that session entirely.
Memory in ChatGPT is personal. A teammate using their own ChatGPT account does not share your memory notes or your project files unless you explicitly export and share them. The assistant that helped you this morning does not know what your colleague asked it last week.
What carries over in Claude
Claude does not maintain persistent user memory across conversations in the same way ChatGPT does. Instead, continuity in Claude comes from Projects: a project holds a set of instructions and uploaded files that Claude reads at the start of every conversation inside it. You write those instructions; they do not update automatically based on what you tell Claude.
Claude Code reads AGENTS.md files from the repository it is working in, which provides a form of project memory for coding work. Claude Code Projects added shared memory in September 2026, allowing team members to share project context within Claude. For comparing Claude and ChatGPT on coding tasks specifically, see Claude vs ChatGPT for coding.
How Gemini handles memory and workspace context
Gemini can reference items in your Google Workspace when you give it access: emails, documents, calendar events. That context is not stored memory in the traditional sense; Gemini retrieves it at the time you ask. The scope depends on the version of Gemini you use and the permissions you have granted.
Gemini Gems — saved AI personas with custom instructions — are becoming skills on personal accounts from November 2026 per Google's announcements. For Gemini's current memory and workspace features, check Google's Gemini help center (opens in a new tab) directly, since these capabilities are updating frequently.
Comparing assistants by what carries over
The table below summarizes the key dimensions. Check each vendor's current help pages before relying on a specific behavior, since all of these products update regularly.
| Assistant | What carries over | Who controls it | Works for teammates? |
|---|---|---|---|
| ChatGPT | User preferences and facts via memory; project instructions and files in Projects. | User can view, edit and delete memory entries; project content is user-managed. | No. Memory is per-account. Projects are not shared across accounts by default. |
| Claude | Project instructions and uploaded files written by the user; AGENTS.md for coding work; shared memory in Claude Code Projects. | User writes and edits instructions; they do not self-update. | Claude Code Projects allow shared context within a team; standard projects are per-account. |
| Gemini | Workspace documents, emails and calendar items retrieved on request; saved instructions in Gems or skills. | Permissions granted by the user; Google Workspace admin controls for business accounts. | Shared Workspace items are accessible; personal memory is per-account. |
| Microsoft Copilot | Microsoft 365 content retrieved from Graph; conversation history within the session. | Microsoft 365 admin settings and per-user permissions. | Shared organizational content in Microsoft 365 is accessible to permitted users. |
| fenbs | Projects, tasks, Decisions and rules, lessons and Where we left off for the whole team — outside any single chat app. | People manage the record; roles set who can read, write and comment; History records every change. | Yes. Team boards give every member's connected assistant the same record, with access controlled by role. |
The common gap across all personal memory systems is that they stop at the account boundary. They do not carry decisions across people, survive a change of AI app, or give a newcomer the context they need for a shared project.
Keep the work when you change AI apps
fenbs holds your projects, tasks, Decisions and rules and Where we left off outside any single assistant. Claude, ChatGPT, Cursor and other connected apps read the same record over MCP. When you switch models or hit a limit, the next session starts from the same project context without you rebuilding the explanation. Connect from Integrations.
Create a free accountFree for one person · No credit card · Works with Claude, ChatGPT, Cursor and more
What personal memory leaves out
Personal memory in ChatGPT, Claude or Gemini stores information about you and your preferences. It does not track the state of active projects, record decisions that teammates made, or tell a different assistant where the work currently stands. A teammate with their own ChatGPT account cannot read your memory notes. An assistant that has never seen a project cannot know its rules and current tasks from memory alone.
This matters when work spans multiple people, multiple sessions or multiple AI apps. Each of those boundaries is a point where personal memory breaks down. That is not a flaw in the tools; it is a difference in scope. Personal memory is designed for preferences and repeating facts about one person, not for shared work records. The AI memory for teams guide explains the distinction in more depth.
fenbs does not connect to your chat history or import memories from ChatGPT or Claude automatically. You build the shared record by adding projects, tasks and decisions, and by asking connected assistants to save progress. The record is what the team and its assistants choose to write there, not a transcript.
How to do this with fenbs
- Create a free account. A personal kanban board is free with no credit card. When teammates need access, create a team board and give members roles.
- Add the project. Write the goal, list the current tasks and record any agreed decisions or rules. Aim for the minimum a new assistant needs to start useful work.
- Connect your AI apps from Integrations. MCP lets an assistant read and update fenbs without you copying and pasting. Set up the apps you use most first.
- At the start of each session, ask the assistant to read the project context: Rules first, then Where we left off, In progress and the current tasks.
- At the end of the session, ask the assistant to save progress: what it completed, what it checked, what remains and the next step. This is what the next session — in any connected app — will read.
When a colleague's assistant starts the same project, it reads the same record. When you switch from ChatGPT to Claude for a session, both assistants work from the same decisions and current tasks. The portable memory guide at portable AI memory covers this pattern in more detail.