Run multiple Claude Code sessions at once. Each on its own branch. None colliding.

The problem with running Claude Code in separate terminals

Running two Claude Code sessions in the same working directory is asking for trouble. Both agents read the same files and write to the same paths. When they conflict, you are untangling a merge mess in the middle of an automated run.

The workaround — running sessions one at a time — erases the benefit. Claude Code is fast, but your time is finite. Sequential sessions waste the most valuable resource: your attention is in one place while Claude Code works in one other place, instead of five.

How Tempest isolates Claude Code sessions

Tempest creates a dedicated git worktree for each Claude Code session. Each worktree is a separate working directory linked to your repository, with its own branch. Claude Code in tab 1 and Claude Code in tab 2 are in completely separate directories. They cannot read each other's in-progress changes, and they cannot produce merge conflicts.

Creating, managing, and cleaning up worktrees is handled automatically. Open a session tab — worktree created. Close it — worktree persists. The branch is ready to review and merge whenever you are.

Token Intelligence for Claude Code

Every Claude Code session reads your codebase to understand what it's working with. In five parallel sessions, that foundational reading happens five times — you pay for each one. Token Intelligence changes this.

When you enable Token Intelligence, Tempest writes an Atlas MCP configuration file for Claude Code automatically. Claude Code picks up the Atlas MCP server and queries the local code-knowledge graph instead of reading files from scratch. Symbol definitions, call chains, import trees — all answered from the graph in milliseconds, at a fraction of the token cost.

Session continuity for Claude Code

Claude Code sessions in Tempest persist when you close the tab. The full conversation history is saved, the branch is unchanged, and the worktree is exactly as Claude Code left it. Reopen the tab and Claude Code picks up where it stopped — no re-orientation, no re-reading the task.

This matters when you are running multiple sessions and reviewing them one by one. Each session you return to is already in context. You review the diff, ship or discard, and move to the next.

What this enables in practice

• Run Claude Code on five different features simultaneously — review and ship when each finishes.
• Try three approaches to the same problem in parallel — all isolated, all reviewable, pick the best.
• Run a bug-fix session and a feature session simultaneously — one doesn't block the other.
• Use Token Intelligence to make five sessions cost significantly less than five times one session.

How to get started

Download Tempest, open your project, and click the + tab to start a new session. Select Claude Code. Tempest creates the worktree, starts the session, and Claude Code is ready. Open more tabs for more sessions.

To enable Token Intelligence: go to Settings → Token Intelligence, turn it on, and index your project. Tempest writes the Atlas MCP config for Claude Code automatically. From that point every Claude Code session in that project uses the graph.