# ChatGPT memory full or not saving: move it to a shared wiki

> What to do when ChatGPT memory is full or not saving, how to get your memories out, and how to keep one Markdown memory that ChatGPT, Claude and Codex all read.

Updated 2026-11-09. Canonical: https://cozy.cloud/chatgpt-memory

> When ChatGPT memory is full or not saving, delete or consolidate entries in its memory settings. To share memory with Claude, copy the saved entries into a Markdown file and keep that file in a store both assistants can read. ChatGPT memories are not documented as part of the standard export, so the copy is manual.

## Why ChatGPT memory fills up or stops saving

ChatGPT's pricing page lists memory as "Limited" on Free, included on Go, and "Expanded" on Plus and Pro ([OpenAI](https://chatgpt.com/pricing)). So the first thing to check is your plan. We could not verify the exact capacity or the mechanics of the "memory full" message, because OpenAI's help pages returned errors when we fetched them. Treat any specific number you read elsewhere as unconfirmed.

A third-party comparison also notes that ChatGPT memory holds high-level context, not verbatim text ([MemoryPlugin](https://blog.memoryplugin.com/claude-vs-chatgpt-vs-gemini-memory/)). That is why an entry can feel vague, and why memory that looks lost may never have been saved in the form you expected.

Practical steps inside ChatGPT:

1. Open the memory settings and read the saved entries.
2. Delete duplicates and entries that are out of date.
3. Merge several small entries about the same topic into one clear statement.
4. Check whether memory is switched on for the account and for the chat you are in.
5. If you hit the limit again, move long-lived facts out of ChatGPT memory and into a file you control.

## Can you export ChatGPT memory to Claude

Not automatically. MemoryPlugin reports that ChatGPT memories can't be imported from other assistants and aren't documented as part of the standard export (same source as above). Claude does have an import path. Anthropic made memory and a memory import tool available on the free plan on 2 March 2026, and the import is a paste-in of another chatbot's output, a one-time snapshot rather than sync ([MacRumors](https://macrumors.com/2026/03/02/anthropic-memory-import-tool/)). See [Claude memory](/claude-memory) for how that side works.

The manual route works for either direction:

1. In ChatGPT, ask it to list everything it remembers about you, as plain text.
2. Paste the answer into a Markdown file and edit it by hand. Remove anything wrong or sensitive.
3. Paste the cleaned text into Claude's memory import, or keep it in a shared file store as described below.

The snapshot approach has a limit. The moment you tell either assistant something new, the two memories drift apart, and you repeat the copy.

## Keep one memory both assistants read

The alternative is to stop treating each assistant's built-in memory as the source of truth. Put your durable context in Markdown pages and let each assistant read and write them through MCP.

Cozy is a private wiki and cloud file space for each person and their AI agents. Every account has one hosted MCP endpoint, `https://cozy.cloud/mcp`, for all of its workspaces. ChatGPT and Claude take it as a custom connector, and Codex and Claude Code add it with one command, then sign in with OAuth. Cozy runs no model itself, so you bring your own agent. See [the Cozy MCP server](/mcp-server) and [connecting ChatGPT](/connect/chatgpt).

A caution on ChatGPT specifically. Secondary sources report that write-capable custom MCP connectors are limited to Business, Enterprise and Edu plans, with Plus and Pro read-only, and those sources note that OpenAI's own documents disagree ([Caipi](https://caipi.ai/blog/which-plans-support-mcp-2026)). Check your plan with OpenAI before relying on ChatGPT to write. Cozy's `read` scope works in a read-only client, while `wiki-write` needs a client that can write.

## What you get compared with built-in memory

| Question | ChatGPT memory | A Cozy workspace |
| --- | --- | --- |
| Format | Entries inside ChatGPT | Markdown, text, JSON, CSV files |
| Shared with Claude and Codex | No | Yes, over MCP |
| You can read and edit every page | Through the memory settings | Yes, as files |
| History | Not documented in our sources | Last 50 revisions readable and restorable |
| Setup | None, it is automatic | Create a workspace and connect each client |

ChatGPT memory is automatic with no setup, and Cozy cannot match that. Cozy's search is lexical only, with no embeddings, so recall depends on how clearly the pages are written.

## Moving your memories in

1. Sign in at cozy.cloud with an email code, Google or Apple. Every account gets "My wiki".
2. Paste your cleaned memory text into a new page under `wiki/`, with a short description in the frontmatter.
3. Or push it through the ingest API as a source, using a stable id so a later push updates the same document:

```bash
curl -X POST "https://cozy.cloud/api/wiki/workspaces/$COZY_WORKSPACE/ingest" \
  -H "Authorization: Bearer $COZY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"documents":[{"id":"chatgpt-memory-export","content":"..."}]}'
```

4. The ingest call uses a bearer credential created under Agents. Bearer credentials are shown once, expire within 90 days, and can be limited by scope or to a folder prefix.
5. Connect Claude, Codex or ChatGPT to `https://cozy.cloud/mcp`, then ask the agent to organize the imported text into pages.

Imported sources land in the built-in Inbox by default and are treated as untrusted evidence, so an agent can cite them without them overriding curated pages. For the idea behind this, see [AI memory](/ai-memory).

## When to stay with built-in memory

If you use one assistant, rarely switch tools and your preferences are simple, ChatGPT's built-in memory is less work. Cozy is worth the setup when you use several assistants, want to inspect and restore what they remember, or keep running into the limit.

## Frequently asked questions

### Why is my ChatGPT memory full?

ChatGPT lists memory as limited on Free and expanded on Plus and Pro, so plan matters. We could not verify the exact capacity from OpenAI's help pages, so open the memory settings, delete duplicates and merge related entries.

### Can I export ChatGPT memory to Claude?

Not as a standard export. Ask ChatGPT to list what it remembers, clean the text, then paste it into Claude's memory import, which is a one-time snapshot. See [Claude memory](/claude-memory).

### How do I share one memory between ChatGPT, Claude and Codex?

Keep the memory in Markdown pages in a Cozy workspace and connect each client to https://cozy.cloud/mcp. ChatGPT write access depends on your plan, so check with OpenAI. See [connecting ChatGPT](/connect/chatgpt).

### Does Cozy replace ChatGPT's memory?

No. It gives your assistants a shared, readable store they can all use. Cozy runs no model itself, so you bring your own agent.
