What you’ll know by the end of this check
- Why multi-session is the biggest shift in the redesign
- The mental model for running parallel Claude work without losing your mind
- Three patterns for when to start a new session vs. continue an existing one
The shortest possible answer
Claude Code desktop now supports multiple concurrent sessions in the same window. Each session has its own context, its own conversation, its own progress. You switch between them like tabs.
This is the feature that stops forcing you to pick one workflow and commit your whole context to it.
Why it’s the biggest change
Before multi-session, a Claude Code session was a single-threaded commitment. You started one, and that session ate your entire context. Want to do something else? Either abandon what you’re doing, /clear, and start fresh (losing context), or leave it and start a new session elsewhere (losing flow).
With multi-session, you can have:
- Session A — refactoring the auth module, six files open, context deep
- Session B — writing release notes for the latest ship, different repo
- Session C — answering a quick “hey can you look at this?” from a coworker without nuking A or B
Each stays alive. Each has its own context. You switch between them, pick up where you left off, never lose the thread.
The mental model
Think of sessions as parallel agent workers, each with its own desk. You’re not multitasking yourself — you’re assigning work to different agents who each hold their own context.
Three patterns worth stealing:
1. Long-running + interrupt
Session A is a long refactor. A coworker pings you with “can you check why this test is flaky?” Open session B. Ask it. Get the answer. Close B or leave it open for follow-ups. Return to A exactly where you left off.
2. Parallel explorations
You’re not sure whether to approach a feature as refactor-first or rewrite-first. Start session A with the refactor plan. Start session B with the rewrite plan. See both plans before committing. Kill the one you don’t like.
3. Compartmentalized repos
One session per repo you work in. Each session has the right CLAUDE.md loaded, the right recent edits in context. You’re not constantly re-orienting Claude to which repo you’re in — each session already knows.
When NOT to start a new session
- When the new work relates to the current session. Branch conflicts, bugs introduced by the current refactor, follow-up questions — stay in the same session so Claude has the context.
- When you want to
/clearanyway. Starting fresh with/clearcosts less than maintaining a second session you’ll barely use. - When you’re about to close your laptop. Sessions are in-memory. A session you won’t come back to today is a session you might as well not have.
The session-management discipline
Multi-session makes it easy to accumulate cruft. Three habits:
- Name your sessions. The redesigned desktop lets you rename sessions. Do it. “refactor-auth” beats “Session 3.”
- Close what you’re done with. A session you won’t return to is just visual noise. Close it.
- Don’t hoard. More than 4–5 live sessions and you’re not multiplying leverage, you’re making a tab hoarding problem.
Things to try right now (8 minutes)
- Open Claude Code desktop.
- Start session A in one of your active projects. Have Claude begin a task you can pause mid-flight (e.g. “find every call to the deprecated auth helper”).
- While session A is running, start session B in a different project. Ask a quick unrelated question.
- Switch back to A. Confirm it’s exactly where you left it.
- Rename both sessions to something meaningful. Close session B if you don’t need it further.
The canonical version
Verify behavior against Anthropic’s official release notes.
Ready to verify this check?
You’ve run two parallel sessions and switched between them without losing context. You’ve renamed at least one session. Mark it cleared.