ChatGPT Memory, Custom Instructions and Projects vs Claude Projects: What Each Agent Keeps (2026)

ChatGPT keeps saved memories, chat history, custom instructions for every chat, and Projects with their own files and instructions. Claude keeps memory as topics, account-wide instructions, and Projects with knowledge and instructions; Claude Code reads CLAUDE.md. Each stays inside its own product: no other agent reads it, and teammates see it only through a shared project or a committed file.

I work in ChatGPT, Claude, Claude Code and Codex every day, so I wanted to know exactly what each one remembers and who else can see it. I read OpenAI's and Anthropic's own help pages on 17 October 2026 (links at the end). Here is what they say today, feature by feature, and where the gaps are.

What each agent keeps, at a glance

Checked on 17 October 2026 in OpenAI's and Anthropic's help centers and docs.

The pattern is the same in every row. Each feature makes one product better at knowing you. None of them is a place your other agents can read.

ChatGPT memory vs custom instructions

OpenAI's own one-line answer: custom instructions are direct guidance you write about what ChatGPT should know and how it should respond, while memory uses information that develops across conversations and other sources.

Custom instructions are a text box. You find them in Settings, Personalization (Customize ChatGPT on mobile). They apply to all chats immediately, on every plan. Free and Go accounts get 1,500 characters; Plus, Pro, Business, Enterprise and Edu get 5,000. They aren't shown to people who open a chat you share by link. Two details matter if you use other tools:

• OpenAI's page warns that if you use third-party plug-ins, the model may pass relevant information from your instructions to the plug-in developer.
• OpenAI says there will be no API for custom instructions. Developers are pointed to system messages in the API instead.

Memory is several things under one setting (Settings, Personalization, Memory):

• Saved memories: details you ask ChatGPT to remember, or that it saves as useful context. They're stored separately from chat history, so deleting the original chat doesn't delete the memory. Depending on plan and platform, you can search, sort and see version history.
• Reference chat history: ChatGPT uses relevant parts of past chats. OpenAI says there is no separate storage limit for it. Turn it off and what it learned from past chats is scheduled for deletion within 30 days.
• Improved memory: a broader summary that ChatGPT keeps updating. You can read a memory summary, correct it, or choose "Don't mention this again".
• Other sources: depending on plan and region, memory can also draw on custom instructions, files in Library and connected apps such as Gmail. When it does, Sources may appear under the answer.

Temporary chats don't create or update memories. Workspace owners and admins control which memory settings members can use.

Which to use: custom instructions for rules you want in every chat, word for word. Memory for the things ChatGPT should pick up as you go. If something must be followed exactly, write it as an instruction; memory decides for itself what is relevant.

ChatGPT Projects

A ChatGPT project holds chats, files and instructions together. Project instructions apply only inside that project and override your global custom instructions. You can also paste Google Drive and Slack links as project sources, save a good answer back to the project, and use connected apps in project chats.

• Files: 5 per project on Free, 25 on Go and Plus, 40 on Pro, Business, Enterprise and Edu, uploaded up to 10 at a time. Projects themselves are unlimited.
• Sharing: available on Free, Go, Plus, Pro, Business, Enterprise and Edu. Members get chat access or edit access. Collaborators are capped at 5 on Free, 10 on Plus and Go, and 100 on Pro and in workspace projects.
• Memory: a project uses default memory or project-only memory. Shared projects always use project-only memory: they can't see any member's personal memories, custom instructions or chats outside the project. You can't list a project's memories; to stop ChatGPT using a chat, you delete or move it.
• Enterprise and Edu: inside a project, your custom instructions don't apply at all, only the project instructions.

Of everything in this article, a shared ChatGPT project is the closest thing to shared team context. It's still a ChatGPT feature: your teammates read it in ChatGPT, and nothing outside ChatGPT reads it.

Claude Projects

Claude's Projects are self-contained workspaces with their own chat history and project knowledge: documents, text and code you upload, which Claude uses in every chat in that project. Project instructions tell Claude how to behave there.

• Who gets them: every plan, including Free, which is capped at 5 projects.
• Capacity: on Pro, Max, Team and Enterprise, when project knowledge nears the context limit, Claude switches to retrieval (RAG) and can hold up to 10 times more.
• Sharing: Team and Enterprise only. Members get "Can view" or "Can edit", and a project can be shared with the whole organization. Owners can turn sharing off.
• Between chats: Anthropic notes that context isn't shared across chats in a project unless you add it to the project knowledge. The project's name and description aren't visible to Claude either.

Anthropic is also rolling out a new version of projects in beta, starting in Claude Code for some Pro and Max users. There, a project is one conversation that Claude splits into parallel threads running in the cloud, each starting with the project's files, repositories, instructions and memory.

Claude memory

Claude's memory is on by default for Free, Pro and Max, on the web, desktop and mobile. On Team and Enterprise, an owner turns it on for the organization, and each member then turns it on for themselves.

• How it's stored: as individual topics, saved while you chat. You can read, edit and delete each topic in Settings, Memory.
• Projects: each project has its own memory. In a project, Claude uses that project's memory plus your account memory, never another project's. Turn off "Use account memory" on the project page to keep it fully separate.
• Cowork: memory is shared between chat and Claude Cowork when Cowork runs in the cloud.
• Search past chats: a separate tool on paid plans; Claude searches your earlier chats when you ask.
• Teams: memory is personal. Owners can't view or edit a member's memories, and turning memory off for the organization deletes everyone's memory permanently.

Claude also has Instructions for Claude, an account-wide setting applied to every conversation: the counterpart of ChatGPT's custom instructions.

Claude Code: CLAUDE.md and auto memory

Claude Code has two memory systems, both loaded at the start of every session:

• CLAUDE.md files are written by you: an organization file, your user file in ~/.claude/, and project files in the repository. Commit the project file and everyone who clones the repository gets it. Anthropic suggests keeping each file under 200 lines. Since v2.1.277 Claude Code also reads AGENTS.md; I covered the details in AGENTS.md vs CLAUDE.md.
• Auto memory is written by Claude Code: your corrections and preferences, saved in ~/.claude/projects/<project>/memory/. The first 200 lines or 25KB of its MEMORY.md index load into each session. Anthropic is explicit: auto memory is machine-local and isn't shared across machines or cloud environments.

Claude Projects vs ChatGPT Projects

Checked on 17 October 2026. Both are good at keeping one stream of work together in one app. ChatGPT lets people on any plan, Free included, share a project; Claude adds retrieval for large knowledge bases on paid plans.

ChatGPT memory vs Claude memory

Both remember you between chats, both let you see and edit what they keep, and both have a way to chat without memory (ChatGPT's temporary chat, Claude's incognito chat or memory off for one chat). The differences: ChatGPT's memory can also draw on Library files and connected apps; Claude stores memory as editable topics and shares it with Cowork in the cloud.

What they share is the boundary. Moving memory between them is a copy and paste: Anthropic's import flow gives you a prompt to run in your old assistant, then you paste its answer into Claude, and Anthropic calls the import experimental. That's a one-time snapshot, not a link. The next thing you teach ChatGPT, Claude won't know.

The gap: every agent keeps its own copy

Put the facts together and you get four separate stores for one business:

1. ChatGPT knows what you told ChatGPT, through memory, custom instructions and its projects.
2. Claude knows what you told Claude, through memory, instructions and its projects.
3. Claude Code knows what is in the repository's CLAUDE.md and what it saved on this one machine.
4. Codex, Cursor and cloud agents know what is in their own config and the files they open.

Your positioning, prices, brand voice and last week's decisions end up written four times, in four places, each slightly different. Teammates see a ChatGPT project only in ChatGPT and a Claude project only on a Claude Team or Enterprise plan. When something changes, you update it in every place by hand, or one agent works from old facts.

Where LotsNotes fits

I built LotsNotes for this gap. You write your business context once, as notes in a workspace with folders, and every agent reads the same current notes over MCP or the REST API.

• One source for every agent. Claude, ChatGPT, Claude Code, Codex and Cursor connect to https://api.lotsnotes.com/mcp and sign in with OAuth; cloud agents and scripts use the REST API with a key. In ChatGPT, adding a custom connector with full read and write depends on your ChatGPT plan, and on some plans only admins can add one; check OpenAI's current plan details. Setup for each client is on the connect page, and I walked through it in how to connect Claude, ChatGPT, Codex and Cursor to LotsNotes.
• AGENTS.md notes. A note titled AGENTS.md holds the instructions agents read first. At workspace level it comes back with every folder's notes, so it reaches every agent; a folder can hold its own for one product, client or agent.
• Private notes never reach agents. Mark a note private and it stays on lotsnotes.com: it's never returned through MCP, the API or any sync. Shared notes follow workspace roles (owner, admin, member, viewer).
• You can read and edit it directly. Every save keeps a version you can restore, and every note an agent reads comes back with its lotsnotes.com link, so you can check what it used.
• Teammates without a per-seat bill. LotsNotes is free for one person: 1 workspace, 3 folders, unlimited notes and checklists, MCP and REST. LotsNotes Pro is $99, paid once, for unlimited workspaces, folders and team members, and it's included while you have a Pro or Business plan of LotsSocial, LotsBlog, LotsTeam, LotsCRM, LotsInvoice, LotsBusiness or LotsStore. Details on the pricing page.

This is how LotsTech runs. My agents' shared rules are one workspace AGENTS.md note, and each agent has a folder with its own AGENTS.md, plan, handoff and memory notes. Claude Code, Codex and the cloud agents all read the same notes, so when I change a rule on lotsnotes.com, every agent has it on its next read.

How I'd split it

You don't have to choose. Keep each vendor's memory for what is personal to that app, and keep shared business facts in one place.

• ChatGPT custom instructions and Instructions for Claude: your personal style. One line on top: "For anything about my business, read the AGENTS.md note in LotsNotes first."
• Project instructions (ChatGPT or Claude): a short pointer to the LotsNotes folder for that product or client, plus anything only that project needs.
• CLAUDE.md and AGENTS.md in a repository: build commands and code rules, plus a pointer to the workspace and folder in LotsNotes.
• LotsNotes: positioning, customers, voice, prices, decisions and each agent's playbook.

A prompt that works in any connected agent:

Use LotsNotes. Find the context for the "Acme" workspace, "Marketing" folder. Read the AGENTS.md notes first, then the brand voice note. Draft three launch emails in that voice. If you learn something new about our customers, add it to the customer insights note and give me its link.

Behind that prompt the agent calls find_context, read_note and save_context_note, the same tools in every client. If you're weighing other places to keep this context, I compared LotsNotes vs Notion for exactly this job, and the rest are on the comparisons page.

FAQ

What is the difference between ChatGPT memory and custom instructions?

Custom instructions are text you write that ChatGPT applies to every chat, up to 1,500 characters on Free and Go and 5,000 on paid plans. Memory is what ChatGPT picks up over time: saved memories, past chats and, depending on plan, Library files and connected apps. Use instructions for rules, memory for context.

Is Claude's memory the same as ChatGPT's memory?

They do the same job differently. Claude saves memory as editable topics, keeps a separate memory per project and shares memory with Cowork in the cloud. ChatGPT combines saved memories, chat history and other sources into a summary you can correct. Both are personal to your account and stay inside their own app.

Claude Projects vs ChatGPT Projects: which is better for a team?

It depends on your plan. ChatGPT lets you share a project on every plan, including Free, with chat or edit access, and shared projects keep members' personal memories out. Claude shares projects on Team and Enterprise, and paid plans add retrieval for large project knowledge. Either way, teammates read the project only inside that app.

Can I move my ChatGPT memory to Claude?

Yes, as a one-time copy. Claude's import flow in Settings, Memory gives you a prompt to run in ChatGPT, and you paste the answer into Claude. Anthropic calls memory import experimental. Later changes don't sync, so the two memories drift apart again.

Can Codex, Cursor or my other agents read my ChatGPT or Claude memory?

Not directly. Memory, custom instructions and project knowledge are used inside ChatGPT or Claude. OpenAI says there will be no API for custom instructions, and Claude Code's auto memory stays on one machine. To give every agent the same context, keep it somewhere they can all read, such as LotsNotes over MCP or REST.

Can my AI agents read my private notes in LotsNotes?

No. Private notes stay on lotsnotes.com and are never returned through MCP, the API or any sync. Agents see only shared notes, within your workspace roles.

Sources

All opened on 17 October 2026:

• OpenAI, Memory in ChatGPT
• OpenAI, ChatGPT Custom Instructions
• OpenAI, Projects in ChatGPT
• OpenAI, Connecting custom MCP servers in ChatGPT
• Anthropic, What are projects?
• Anthropic, How can I create and manage projects?
• Anthropic, Use Claude's chat search and memory to build on previous context
• Anthropic, Understanding Claude's personalization features
• Anthropic, Import and export your memory from Claude
• Anthropic, How Claude remembers your project (Claude Code)

Write it once

Let ChatGPT and Claude remember your preferences. Put the business context every agent needs where every agent can read it: connect your agent to LotsNotes and start with one AGENTS.md note.