Open-source app Tempest runs multiple AI coding agents side by side
Tempest is an lication built to help with . Rather than relying only on larger (the amount of text an AI can consider at once), it gives AI agents a local code graph — a structured map of how a codebase's files and pieces relate to each other — so they can understand a repository more effectively. In internal benchmarks, this cut token usage by 64%, meaning agents spend fewer resources searching for relevant context and more on actually solving problems.
Tempest lets a single interface run several at once, including Claude Code, Codex, Gemini CLI, , Cursor Agent, Cline, Goose, and OpenCode. Each agent runs in its own isolated environment, using separate Git worktrees plus OS-level process isolation — Bubblewrap on Linux, Seatbelt on macOS, and a custom isolation layer on Windows — so multiple agents can execute commands in parallel without interfering with each other. Everything runs locally, there is no vendor lock-in, and the entire project is open source.
The team also builds Tempest using Tempest itself, testing every new feature through the same they are building for others.
Key points
- Uses a local code graph to help agents understand repos, cutting token usage by 64% in their benchmarks
- Runs multiple agents — Claude Code, Codex, Gemini CLI, Cursor Agent, and more — from one interface
- Combines Git worktrees with OS-level process isolation (Bubblewrap, Seatbelt, custom Windows layer) for safe
- Runs entirely locally, no vendor lock-in, fully open source