A 7MB IDE built to manage AI agents, not humans

What it does Nezha is a Tauri-based desktop application that wraps Claude Code and Codex in a multi-project workspace. It spins up native agent sessions inside a built-in terminal, tracks their task lifecycles, and bundles a lightweight code editor, Git client, and usage dashboard so you can delegate, review, and commit without leaving the app. The pitch is simple: treat the AI as the primary user of the IDE, and the human as the supervisor switching between contexts.

The interesting bit The app auto-discovers agent sessions and highlights when one stalls waiting for your input, which is the exact failure mode of parallel vibe coding. It also claims a 7MB install package—smaller than most VS Code extensions—yet still manages to squeeze in branch management, diff views, and AI-generated commit messages.

Key highlights

• Runs multiple Claude Code / Codex instances simultaneously across projects with one-click context switching
• Built-in terminal emulator (xterm.js) that pipes real-time agent output and preserves interactivity
• Session auto-discovery and playback: resume interrupted tasks by browsing visualized history
• Native Git integration including diff views, branch management, and AI-assisted commit message generation
• Weekly token and tool-invocation analytics to track how much your agents are actually costing you
• Light and dark themes, because even headless agents deserve ambiance

Caveats

• Requires Claude Code and Codex to be pre-installed separately; it is a wrapper, not a standalone agent
• macOS builds are currently unsigned, so Gatekeeper will flag the app as damaged on first launch
• The code editor is intentionally lightweight (Shiki and CodeMirror), so heavy refactoring or complex debugging likely still wants a traditional IDE

Verdict Grab it if you are already juggling multiple AI coding sessions and need a minimal control tower. Stick with your current setup if you do most of the typing yourself or need a full-featured debugger.