Run multiple Claude Code sessions at once

Can you run more than one Claude Code session at once?

Yes. Open another terminal in your project and run claude. Each one is its own session with its own conversation. Anthropic’s docs describe exactly this: run claude --worktree again with a different name in a second terminal, and you have a second isolated session.

The commands are easy. Three things go wrong after that: sessions overwrite each other’s files, you lose track of which one is waiting for you, and your usage limit drains faster than you expect. Each has a fix.

How do you stop sessions from overwriting each other?

Give every session its own copy of the project. Two agents editing one folder will rewrite each other’s changes, and a test that fails because of the other agent’s half-finished edit tells you nothing. OpenAI’s Codex docs say the same: run chats concurrently, but avoid letting two of them change the same files, and use worktrees for separate checkouts.

A git worktree is a second working directory with its own branch, sharing the same history. Claude Code creates one for you:

claude --worktree feature-auth

By default that makes .claude/worktrees/feature-auth/ on a new branch named worktree-feature-auth, per the worktrees docs. Add .claude/worktrees/ to your .gitignore. A worktree is a fresh checkout, so install dependencies in it, and list gitignored files such as .env in a .worktreeinclude file to have them copied in automatically. Our guide to git worktrees for AI agents has the full setup.

How do you keep track of which session needs you?

This is the part that costs real time. A session that stops at a permission prompt can sit there while you watch another one. You have four options:

A terminal multiplexer. tmux lets you create, access and control several terminals from a single screen, and it can detach and keep running in the background. It puts every session on one screen. It does not know which agent is waiting for you.

Notifications. Claude Code fires a notification event when it finishes a task or pauses for permission and you appear to be away. By default it sends a desktop notification only in Ghostty, Kitty and iTerm2. In other terminals, set preferredNotifChannel to terminal_bell, or add a Notification hook that plays a sound or runs a command. The hooks guide has copy-paste examples for macOS, Linux and Windows.

Agent view. As of October 2026, claude agents opens one screen listing your background sessions with a state for each: working, needs input, idle, completed, failed or stopped. Start one with claude --bg "task". Anthropic labels it a research preview, so expect the interface to change. See the agent view docs.

The desktop app. Anthropic’s Claude desktop app has a New session button for parallel tasks, with a worktree option so each session gets its own copy.

How many sessions can you actually supervise?

Fewer than you think. Every session produces a diff that you have to read. When you stop reading carefully, you are not supervising any more. You are approving.

Start with two sessions on two unrelated tasks. Add a third only after you can review the first two without rushing. That is a rule of thumb, not a measurement, so adjust it to your own review speed.

Usage is the other limit. The agent view docs say that running ten agents in parallel uses quota roughly ten times as fast as running one. On a subscription, that multiplier shows up as reaching your five-hour limit sooner, so plan the number of sessions around your plan, not your screen.

Give each session one task with a check it can run, such as a test command, so you know when it is done without reading every line of its output.

What does a good setup look like?

Keep it boring. One repository, one terminal tab per task, each tab named after its branch. Every task runs in its own worktree. Every prompt ends with the command that proves the task is done, such as npm test.

Leave one tab free for yourself. That is where you run git status, read the diff and run the checks before anything is merged.

Merge one branch at a time. After each merge, update the next branch from your main branch and run its checks again, because a change that passed alone can fail next to another.

What if some of your agents are not Claude Code?

Anthropic’s parallel tools run Claude sessions. Its docs say that in every approach the workers are Claude sessions, and that to involve a different tool you expose it to Claude as an MCP server.

If you also use Codex, Gemini CLI or Cursor’s CLI, each has its own way to isolate work. Gemini CLI has an experimental --worktree flag, Cursor’s agent --worktree creates worktrees under ~/.cursor/worktrees/, and Codex offers worktrees in the ChatGPT desktop app. Each reports its own state in its own window. With a mixed set of agents you end up with several windows and no single list of what needs you.

See every agent in one window

SwarmPane’s Agent Grid runs up to 16 agents in real terminals, each pane showing whether it is working, waiting for you or done. Status comes from each CLI’s own signals: Claude Code hooks, terminal notifications and exit codes. A CLI that reports nothing shows “limited status” instead of a guess.

The Attention list orders every session across your projects, the ones waiting for you first. Layouts can open a launch set of agents in one click, each in its own worktree if you want.

SwarmPane runs the CLIs and accounts you already have. Start with a 7-day trial for $1 and put your next three sessions side by side.