Choose how to run several Claude Code sessions at once.

Start with what Claude Code includes

Subagents run inside one session, each with its own context, and return a summarized result to the session that called them. Agent teams go further: one session acts as the lead and coordinates teammates, separate Claude Code instances that message each other and share a task list. Agent teams are experimental and off by default; you enable them with CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1.

Agent teams add coordination overhead and use significantly more tokens than a single session, because each teammate is its own Claude instance. They suit independent work, such as parallel research, separate modules or competing hypotheses. For sequential steps or edits to the same files, a single session or subagents usually fit better.

• Subagents: results return to one session, at a lower token cost.
• Agent teams: teammates message each other directly; experimental, with a higher token cost.

Give every session its own working copy

Two sessions editing the same checkout can overwrite each other’s changes. Git worktrees give each session its own working directory on its own branch, so changes stay separate until you merge them. Most orchestration tools build on this pattern.

Remote Control’s server mode applies it too: started with claude remote-control --spawn worktree, each session you open from another device gets its own git worktree, with up to 32 concurrent sessions by default. You still need an integration step: a person, or a final session, has to combine the branches and check the result as one change.

Compare desktop apps, terminal managers and boards

Desktop apps such as Conductor, for Mac, run Claude Code, Codex and other agents in parallel, and give each task its own workspace, branch, terminal and diff. Terminal managers such as Claude Squad run each agent in its own tmux session and git worktree, for people who work in the terminal. Task boards such as Vibe Kanban plan the work as issues on a board and start workspaces where an agent works on its own branch, so you can review the diff before opening a pull request.

Hosted options exist as well. In February 2026 GitHub added Claude and Codex as coding agents in Agent HQ, in public preview for Copilot Pro, Pro+, Business and Enterprise plans, and Codex cloud runs coding tasks in parallel cloud environments. These tools change quickly: in April 2026, for example, the makers of Vibe Kanban announced that it would continue as a community-maintained open-source project. Check the current documentation, licence and maintenance status before you adopt one.

• Where do sessions run: your machine, your server or a provider’s cloud?
• How is each session kept separate, and who merges the results?
• Can you follow and steer sessions from another device?
• Can other people see the work and review it?

Use a shared workspace when the work involves people

Most session managers are built for one developer at one computer. When requests come from colleagues, several people review results or agents run on more than one machine, the problem changes: the brief, the sessions, the decisions and the review need to live in one place that everyone can see.

TM8 is built for that case. People and Claude Code or Codex sessions work in one shared workspace: tasks and documents carry the context, sessions run in parallel on a hosted seat or on your own machine or server, a session can work on its own branch in a separate worktree, requests can arrive from WhatsApp and results come back for human review. If the whole team is one developer at one desk, a lighter tool may be all you need.

Common questions