What is the best way to manage multiple Claude Code sessions?

Give every session a name and its own git worktree, then watch them from one place that tells you which one is waiting. For background sessions, Claude Code's own agent view (claude agents) does this. For interactive sessions you drive in terminal tabs, you need a terminal setup or an app that surfaces the waiting state. The best choice depends on whether you prefer to dispatch tasks or to steer each conversation.

What "managing" sessions actually means

Past two or three sessions, the same four problems show up:

1. Finding a session again. Which tab was the auth refactor?
2. Knowing which one is blocked. One session waits on a permission prompt while you read another.
3. Keeping their edits apart. Two sessions in one checkout edit the same files.
4. Watching usage. Every session draws from the same plan limits.

Any setup worth keeping answers all four.

Without Kanbell: the native toolkit

Name and isolate every session

claude -w auth-refactor -n "auth refactor"
claude -w billing-fix -n "billing rounding"

-w (--worktree) puts the session in .claude/worktrees/<name>/ on its own branch. -n (--name) sets the name shown in /resume and in the terminal title. Reopen a session later with:

claude --resume "auth refactor"

Inside a session, /rename changes the name and /color tints the prompt bar.

Use agent view for work you hand off

As of October 2026, agent view is a research preview:

claude agents

It lists background sessions grouped by state: needs input, working, completed, failed. Space peeks at a row and lets you reply; Enter attaches to the full conversation. To send a session you already have open into this list, run /bg in it.

By default it shows background sessions from all your projects. To narrow it:

claude agents --cwd ~/projects/my-app

While agent view is open, Claude Code also sends a notification when a background session needs input, finishes or fails.

Check usage

/usage shows session cost, plan usage limits and activity stats. On Pro, Max, Team and Enterprise plans it breaks down what counts against your limits.

Other options (as of October 2026)

• Claude Code Desktop. The Code tab lists sessions in a sidebar, can show two side by side, and can start each in its own worktree. It sends an OS notification when a session finishes and you aren't viewing it.
• Claude Squad (open source). A terminal app that runs Claude Code, Codex, Gemini, Aider and other local agents in separate workspaces. It needs tmux and gh.
• cmux (open source, macOS). A terminal built on libghostty. Panes get a ring and tabs light up when an agent needs attention; a sidebar shows branch and PR status.

Limits of the native toolkit

• Two worlds. Agent view tracks background sessions. Interactive sessions in your terminal tabs don't appear there until you background them. If you like to steer some sessions by hand, half of your work is outside the overview.
• No project grouping for interactive tabs. Ten tabs across three repositories look alike. Names and colors help, but you sort them in your head.
• Waiting signals depend on your terminal. By default, desktop notifications only work in Ghostty, Kitty and iTerm2, and only when you seem to be away. Inside tmux, they need allow-passthrough.
• One vendor. Agent view only manages Claude Code. Codex CLI or Gemini CLI sessions need their own tracking.

With Kanbell

Kanbell is a macOS app (Apple Silicon) built around the "interactive sessions in real terminals" style, organized by project.

• One window per project. Each window holds real terminals in panes and tabs, so sessions for one repository stay together.
• Agent map. For Claude Code, Kanbell maps agents and their subagents with their state: working, waiting or done, and whether the wait is a question or a permission.
• Waiting at a glance. An orange dot in the window, the menu bar and the overview, plus a macOS notification. ⌘⇧J jumps to the next agent that is waiting.
• More than Claude Code. Codex CLI and Gemini CLI show their state and permission waits. Other CLI agents run in the terminals without tracking.
• Usage beside the work. Claude Code subscription usage for the session and the week, in the app.
• Tasks and git. A board (local or GitHub Issues) to launch an agent from a task, and a git panel with changes, diffs, history and worktrees.

What Kanbell doesn't do: it doesn't replace agent view's dispatch-and-forget flow, and it has no Linear, Jira or Trello board. No account and no Kanbell server are involved.

Which setup fits you

• You hand off tasks and check back later: agent view.
• You prefer a chat window to a terminal: Claude Code Desktop.
• You live in tmux and run several vendors: Claude Squad or a tmux layout. See whether you need tmux.
• You drive interactive sessions across several projects and want the waiting state, board and git in one window: Kanbell.