What ChatGPT's memory does today
OpenAI's current help article describes two memory experiences. Improved memory draws on your past chats, saved memories, custom instructions, files in your Library and connected apps such as Gmail, depending on your plan and region, and keeps a continually updated summary. Legacy saved memories keep an explicit list of individual items; where the option exists, you can switch between the two in Settings › Personalization › Memory.
The memory summary is, in OpenAI's words, a high-level view that does not necessarily include every detail Memory can reference. You can correct it, tell it not to mention something again, or delete and turn off memory. Temporary Chats never create or update memories. OpenAI says there is no separate storage limit for what ChatGPT can reference through chat history, and that memory features vary by plan, region, platform and workspace. The article documents no way to export your memory. Source: OpenAI: Memory in ChatGPT.
Can ChatGPT use a memory that lives outside ChatGPT?
Partly. ChatGPT connects to remote MCP servers through developer mode, but OpenAI documents full MCP support, including write actions, only as a beta for Business, Enterprise and Edu workspaces on the web. Pro users can connect MCP servers with read and fetch permissions in developer mode. OpenAI's article does not mention custom MCP for Plus, Go or Free, and MCP apps are not available on mobile. For most individual users today, the practical routes into ChatGPT are a browser extension or a Custom GPT; MCP is the route into Claude, Cursor and other clients. Source: OpenAI: Developer mode and MCP apps in ChatGPT.
1. Claude — the closest replacement
What it is: Anthropic's assistant, with memory on every plan. It is on by default for Free, Pro and Max on web, desktop and mobile; on Team and Enterprise an owner makes it available and each member turns it on. Claude saves memory as individual topics you can open, edit or delete.
Moving from ChatGPT: Claude documents an import flow: you ask your current assistant to write out what it remembers and paste it into Claude. Anthropic labels it experimental and says Claude may not always incorporate imported memories. Claude also connects to remote MCP servers as custom connectors on every plan (Free is limited to one).
Honest limits: it replaces ChatGPT rather than adding memory to it, and the import is experimental. Sources: Claude memory, import and export, custom connectors.
2. Gemini — if you live in Google
What it is: Google's assistant, which personalises answers from your past chats when the feature is on and needs a personal Google Account. Gemini can import memories and full chat-history .zip files from other AI platforms, including a ChatGPT data export.
Honest limits: Google says personalisation features are not available to everyone yet, import requires an adult account, and import is not available in the EEA, Switzerland or the UK. Sources: Gemini personalization, import from other AI platforms.
3. MemoryPlugin — keep ChatGPT, add memory that travels
What it is: one memory shared across many AI tools, including ChatGPT, Claude, Gemini, Perplexity, Claude Code and Cursor. It connects through a browser extension, a Custom GPT for ChatGPT, an MCP server, an API and a macOS app, and lets you export or delete your memory at any time.
Honest limits: no free plan; the 7-day trial asks for a card. Annual plans start at $89 a year (Core) per its homepage on 4 October 2026. Source: memoryplugin.com.
4. Supermemory — memory API with a hosted MCP server
What it is: a memory API, MCP server and plugins, available in the cloud, in your own VPC or self-hosted. The hosted MCP server uses OAuth, so no API key is needed, and its tools both search and add memories. The main repository is open source under MIT.
Price: free plan with $5 of credits a month; Pro is $19 a month. Honest limits: usage is credit-metered, the product reads as developer-first, and the older standalone MCP repository is marked deprecated, so older setup guides are stale. Sources: pricing, MCP docs, GitHub.
5. Mem0 — memory layer for agents you build
What it is: an open-source memory layer (Apache-2.0) available as a library, a self-hosted server or a managed platform, with a hosted MCP server documented for Claude Code, Cursor, Codex, Windsurf and VS Code. Its free tier covers 10,000 memory writes and 1,000 retrievals a month; Starter is $19 a month.
Honest limits: developer infrastructure rather than a consumer app, ChatGPT is not in its MCP client list, and the separate OpenMemory app was removed from the repository in July 2026, so guides recommending it are out of date. Sources: Mem0 MCP, pricing, OpenMemory removal.
6. Recall — summarise what you save
What it is: a save-and-summarise knowledge base with browser extensions, now at recall.it. Its MCP server is documented for Cursor, Claude, ChatGPT (as a custom MCP app), Perplexity and Gemini CLI. Free plan includes 10 AI summaries a month and MCP access; Plus is $10 a month billed yearly.
Honest limits: Recall says its MCP server is currently read-only, so an assistant can look things up but cannot save into it. Sources: Recall MCP, pricing.
7. BrainTube — one library every AI can read
What it is: BrainTube is ours, so read this entry as labelled. It compiles the YouTube videos, podcasts, articles, PDFs and notes you save into one knowledge base, served from a hosted MCP endpoint (https://mcp.brain-tube.com/mcp) with nothing to install. It works with Claude.ai, Claude Desktop, Cursor, Perplexity, VS Code, Gemini CLI and Codex CLI. It is read-write: assistants can search it and save notes and content back into it. MCP is included on every plan, including Free (30 credits a month); paid plans start at $9 a month, and your library exports as Markdown.
Honest limits: it remembers what you save, not your chats in the background, so it is not a replacement for ChatGPT's profile-style memory. ChatGPT support (developer mode, or OpenAPI for custom GPTs) is still in testing. Connect guide · pricing.