Where Claude Code stores your session history

Claude Code writes every session to disk as it happens. If you have used it for a while, you have a detailed record of your work in ~/.claude/projects, until the 30-day cleanup removes it.

#Where the files are

~/.claude/
├── projects/
│   └── -Users-you-dev-my-app/          # one folder per working directory
│       ├── 7f3c...e21.jsonl            # one file per session
│       └── 7f3c...e21/
│           ├── subagents/              # transcripts of subagents it spawned
│           └── tool-results/           # large tool outputs, stored apart
├── history.jsonl                       # every prompt you typed, all projects
├── file-history/<session>/             # file snapshots before edits
└── settings.json

The project folder name is the working directory's path with every non-alphanumeric character turned into -. Very long paths are cut at 200 characters with a hash added.

On Windows the same tree lives under %USERPROFILE%\.claude. Setting CLAUDE_CONFIG_DIR moves all of it.

Find your biggest projects:

du -sh ~/.claude/projects/* | sort -h | tail

#What is inside a session file

Each line is one JSON record. Anthropic describes the format as internal and subject to change between versions, so treat field names as a snapshot. Common top-level fields:

Tool calls appear inside assistant messages as tool_use blocks, and their results come back in the next user record. What is inside an agent session file.

#Keep sessions longer

The default cleanup deletes transcripts, file snapshots, plans and debug logs older than 30 days. To keep a year, add this to ~/.claude/settings.json:

{
  "cleanupPeriodDays": 365
}

It takes effect from now on; it does not restore anything already removed. Your prompt history in history.jsonl is not part of the cleanup and stays until you delete it.

#Resume a session

claude --continue            # the most recent session in this folder
claude --resume              # pick from a list
claude --resume <id|name>    # a specific session, or a path to its .jsonl
claude --resume <id> --fork-session   # branch off without changing the original

Inside a session, /resume opens the picker, /rename names the session, and /branch forks it.

#Export a session

/export copies the current conversation to the clipboard or saves it as a text file. For a readable archive of many sessions, tools such as claude-code-log turn the JSONL files into HTML or Markdown. Tools that read agent logs.

#Delete history

claude purge --dry-run       # what would be removed for this project
claude purge                 # remove this project's sessions
claude purge --all           # every project

Before version 2.1.288 the command was claude project purge. Deleting the files in ~/.claude/projects by hand works too. If you use the desktop app or claude.ai/code, those keep their own session lists.

#Privacy notes

• Transcripts are plain text. Anything the agent read - .env files, command output, API responses - is in them. Redact before sharing.
• On Free, Pro and Max plans, Anthropic may train on Claude Code sessions unless you opt out in your privacy settings. Team, Enterprise and API use are excluded. What the terms say.

#Measure it

The evaluator reads ~/.claude/projects in your browser and counts sessions, active hours, tool calls and the products your agents used.