# What Claude Code remembers between sessions, and what it doesn’t

Each new Claude Code session starts with a fresh context window. It loads your CLAUDE.md instructions and, when enabled, its auto memory. Use claude --continue to reopen the latest session in the current directory, claude --resume to choose a saved session, or /resume from inside Claude Code.

Updated 7 November 2026

## At a glance

- **CLAUDE.md:** Instructions you write. Loaded in every session.
- **Auto memory:** Notes Claude writes itself, on one computer, for one repository.
- **Not carried over:** A fresh session starts without the previous conversation history.
- **A note:** What you want the next session, and you, to read tomorrow.

## What each one keeps

Claude Code’s side read on 7 November 2026 in its documentation, at code.claude.com/docs/en/memory.

| | CLAUDE.md | Auto memory | A note in ThreadPage |
|---|---|---|---|
| Who writes it | You | Claude | You, or a session when you or your rule ask |
| When it’s read | Every session | Every session: the first 200 lines or 25 KB | When you or your rule ask, with Pro |
| Where it lives | Your project, your user folder or your organization | On this computer, one folder per repository | A thread of notes, in your browser and, with Plus or Pro, your account |
| On another computer | If it’s in the repository | No | Yes, with Plus or Pro |
| Readable by Codex | Not by default: configure CLAUDE.md as a fallback | No | Yes, with Pro |
| Readable on your phone | Through your repository | No | Yes, with Plus or Pro |

## CLAUDE.md is for rules

Put in it what every session should know: how to run the tests, what not to touch, how you name things. Claude Code’s documentation suggests staying under 200 lines per file, and it reads the file as context, so a rule can still be skipped.

## Auto memory is what Claude noticed

Claude saves what it learns from your corrections to a folder at ~/.claude/projects/<project>/memory/. All worktrees of one repository share it, it stays on that computer, and you can read or delete it with /memory.

## What to put in a handoff note

Auto memory can retain selected project context and decisions. A handoff note lets you review and share where the work stopped, what you ruled out and what to try next.

## A handoff note to copy

Ask the session to write this before it stops, or write it yourself.

Handoff note:

```
Done:
Decided (and why):
Tried and failed:
Stopped at:
Next:
```

## One thread that Claude Code and Codex both read

When you write to a session with @ in the thread, it also reads what your other AIs wrote there. So “@codex carry on” starts from Claude Code’s last note, with nothing pasted.

## Questions

### Does Claude Code have memory across sessions?

Yes. Claude Code loads your CLAUDE.md instructions and, when enabled, its auto memory. You can reopen a saved conversation with claude --continue, claude --resume or /resume.

### Can Codex read what Claude Code wrote?

Codex reads AGENTS.md by default. You can add CLAUDE.md to project_doc_fallback_filenames. At the start of a run, Codex builds the project instruction chain from the project root (usually the repository root) to its current working directory, with at most one non-empty file per folder. AGENTS.override.md and AGENTS.md take priority over fallback files. With Pro and Sync on, Claude Code and Codex can also read and add to the same ThreadPage notes.

### Does Claude Code’s memory reach my other computer?

Auto memory doesn’t: its files stay on the computer that wrote them. CLAUDE.md does if it’s in your repository.

## Limits

- ThreadPage doesn’t record your sessions. A session reads and adds notes through the ThreadPage plugin.
- Reading notes needs Pro with Sync on. On Free and Plus, an agent can add notes to your Inbox but can’t read them.
- Under @, a session reads what your own AIs wrote. Notes from people and scripts it reads only when you add them as a reference.

Claude Code and Codex are the names of their owners. ThreadPage is independent and not affiliated with or endorsed by them.

## Start with one thread.

[Start free trial](https://thread.page/auth/sign-up?returnTo=%2Fapp%2Fsettings%2Fconnections%2Fcoding-agents%23trial)

14 days of Pro. No card needed. Ends automatically. Choose to continue with Pro for $99.99 a year ($8.33 a month). [See monthly pricing](https://thread.page/pricing.md)

[Start writing](https://thread.page/app)

Read next

### [Hand off from Claude Code to Codex before the limit](https://thread.page/use-cases/agent-work-log.md)

Near the limit in Claude Code? Have the session write where it stopped in a note. Your Codex session reads it and carries on from there.

Pro · Claude Code and Codex

### [Use Claude Code and Codex together, with shared notes](https://thread.page/use-cases/claude-code-and-codex.md)

Claude Code builds, Codex reviews, and both read the same thread of project notes. The plan, the decisions and every reply stay there after the sessions end, so you don’t explain the project to the next one.

Pro · sessions by name

[Install the plugin](https://thread.page/help/connections/coding-agents.md)

[A notes app with MCP for Claude, ChatGPT and Codex](https://thread.page/notes-mcp.md)

[How to keep a work log](https://thread.page/guides/work-log.md)