HOW IT WORKS
Two primitives. Git worktrees for isolation. Token Intelligence for shared context.
Tempest is a Tauri desktop app — Rust backend, React frontend, native WebView. Everything runs on your machine. No cloud, no servers, no data leaving your environment.
Open a project
Point Tempest at a local git repository. The overview screen shows your recent projects and open sessions. Each project has its own set of agent sessions, Token Intelligence index, and workspace state.
Start an agent session
Click the + tab to open a new session. Pick an agent — Claude Code, Aider, Cline, Goose, or any terminal-based tool. Tempest creates a new git worktree and a fresh branch for that session automatically. The agent starts in its own isolated working directory.
Run agents in parallel
Open as many session tabs as you need. Each runs in its own worktree on its own branch. They cannot interfere — no shared files, no shared state, no merge conflicts mid-run. A live status indicator on each tab shows whether the agent is working or done, without you clicking in.
Token Intelligence indexes your codebase
Atlas (the local code-knowledge graph) runs in the background on first open. It builds a semantic index of every symbol, import, and cross-file relationship in your project. Once indexed, every parallel session queries the graph instead of reading files from scratch — up to 64% fewer tokens across sessions.
Review, commit, push
When an agent finishes, open its diff viewer. Review every changed file line by line. Stage what you want to keep, write a commit message, and push — all without leaving Tempest. Each session has its own branch so you can review and merge changes independently.
Sessions persist between visits
Close a tab and the session is saved exactly as it is — conversation history, branch state, worktree contents. Reopen it and the agent picks up where it left off. Nothing is lost between sessions.
Under the hood