Overview
Tempest is a free, open-source (Apache 2.0) desktop app for ‘agentic engineering’ — running several AI coding agents at once without them colliding. You point it at a git repo and it spins up isolated worktrees, one per agent session, so Claude Code, Codex, Aider, Cline, Gemini CLI and others each work on their own branch with zero merge conflicts. A shared local code-knowledge graph lets agents reuse context instead of each rescanning the repo, which Tempest claims cuts context tokens by up to 86% and tool calls by up to 92%. It is native (Tauri 2), runs fully on your machine, and needs no account — so your code never leaves the laptop. For developers who already lean on CLI agents, it turns ‘run one agent’ into ‘run five and keep them straight.‘
Key Features
- Parallel agents, zero conflicts — each session gets its own git worktree and branch automatically, so five agents can build five features at once without stepping on uncommitted changes.
- Shared local knowledge graph — a code index on your machine is reused across sessions, cutting duplicate context fetching and token spend.
- Built-in diff, commit, push, PR — review each agent’s changes in a live diff viewer and ship without leaving the app.
- Live preview and real terminal — your dev server updates inside Tempest as agents edit, with a full PTY terminal per session.
- OS-level sandbox (Hephaestus) — processes are isolated per platform (Job Objects / sandbox-exec / bubblewrap), so a rogue agent can’t reach your main branch.
- Cross-platform and free — Windows, macOS, Linux; no subscription, no per-seat cost.
Pricing
| Plan | Price | For |
|---|---|---|
| Free (Apache 2.0) | $0 | Everyone — full desktop app, all features, local-first, no account or subscription |
Tempest is free and open source; the team ships new features weekly and may add optional paid, team, or cloud-sync tiers later. Check tempestai.dev for current pricing before assuming any paid plan.
Comparison
vs. Claude Code: Claude Code is a single CLI agent; Tempest orchestrates many of them in parallel with isolation. Use Claude Code for one focused task, Tempest when you want five approaches running at once.
vs. Cursor: Cursor is an AI-native editor for one developer; Tempest is a runner that can host Cursor’s agent plus others side by side. Pick Cursor for daily editing, Tempest for parallel agent experiments across tools.
vs. Aider: Aider pairs with your editor in one terminal; Tempest gives Aider its own sandboxed worktree among siblings. Choose Aider for tight loop editing, Tempest to compare Aider against Codex or Cline on the same repo.