ZUS
AI-native IDE: agents code with human review
Start new thread

ZUS - AI-native IDE: agents code with human review

by
•
ZUS is a desktop IDE with an AI agent built in. The agent reads your codebase, plans, edits and runs commands within the permissions you grant. You read every diff before keeping it, undo what it got wrong, and see each subagent's tool calls in Team Inspector. Your project index and context stay on your machine, and you connect your own AI provider. Built on Tauri 2 and Rust, with a ~20 MB Windows installer. Runs on Windows, macOS (Apple Silicon) and Linux.

Add a comment

Replies

Best
Maker
📌

Hi Product Hunt, I'm Jason, co-founder at 8 Sync. We're a small team in Ho Chi Minh City, and ZUS is the IDE we wanted for ourselves.

Coding with an AI agent took most of our time outside the prompt: explaining the project, watching what the agent touched, reading its diffs and undoing the wrong ones. We did that across an editor, a chat window, a terminal and a Git client. ZUS puts all of it in one window.

What it does today:

  • The agent reads your codebase, plans, edits and runs commands, only within the permissions you grant.

  • Every change shows up as a diff you keep or undo.

  • Team Inspector shows which subagent called which tool, and what is still running.

  • Your project index stays on your machine, and you connect your own AI provider.

  • It's built on Tauri 2 and Rust, so the Windows installer is about 20 MB.

It runs on Windows, macOS (Apple Silicon) and Linux. ZUS is still early, and some features depend on your OS and the provider you connect.

Feedback we'd love: where did the agent surprise you, and could you tell what it changed without digging? If something breaks, post it here and we'll look at it.

congrats!! reviewing every diff works for small stuff but on a big refactor ppl will prob just start approving without reading

The local index + bring-your-own-provider combo is what caught my eye — I build macOS apps and I'm not keen on my codebase leaving the machine. Curious how it handles Swift/Xcode projects, since most agent IDEs feel very JS/TS-first. Does the agent understand Xcode project structure, or is it more of a generic file-based approach?

Maker

 Thanks mate, and that’s exactly the setup ZUS is built for. The index lives on your machine, and the agent talks only to the provider you connect.

To be straight about Swift/Xcode today: it’s mostly a generic, file-based approach.

  • Index: the local index uses real syntax parsing for Rust, TypeScript/JavaScript, Python and Go. Swift files are still indexed and searchable, but they’re split with a simpler text heuristic, not a Swift parser, so retrieval is less precise than for those languages.

  • Xcode projects: the agent doesn’t model .xcodeproj / .pbxproj. It sees your project as folders and files. Swift Package Manager projects fit that model better than Xcode-only ones.

  • Editor: Swift has syntax highlighting. The Swift language extension (SourceKit-LSP) is off by default right now, so you won’t get Xcode-level completion out of the box.

  • Verification: the agent can run your own commands, for example swift build, swift test or xcodebuild test, and ZUS shows the real command output as evidence instead of trusting the agent’s “done”. I haven’t tried this on a full Xcode project myself, so treat it as untested for now.

  • macOS: the build is Apple Silicon only.

So yes, the stack is JS/TS-first today, like you suspected. Proper Swift parsing in the index is a gap we know about. If you’re up for trying it on a real macOS project, feedback on where it breaks would really help us prioritize that.