It runs on any Mac with macOS 14 or later, Intel or Apple Silicon.
People often ask whether it handles real project complexity. It works across multiple repositories. You browse a visual file tree, search, and select exactly what the AI sees. No guesswork.
Another common question is about safety. Every AI-generated change passes through a code review mechanism. Approve edits one by one or in batches. You stay in control.
In the first days, resist the urge to point it at your largest repo. Start with something smaller. A side project. A single utility. Let yourself learn the rhythm of the review panel before you face a hundred changed files.
The prompt library rewards early attention. Save the two or three prompts you reach for repeatedly. Name them well. A week from now, muscle memory will thank you.
Keep an eye on the token counter during your first sessions. It sits in the corner quietly, but it teaches you what a prompt actually costs. You will learn to select files more deliberately.
Option-click previews
Hold Option while clicking a file in the tree. It opens a quick preview without leaving your current selection. Useful when you need to check a function signature without disrupting the context you have already built.
Agent-mode runs can be long. The recent refresh improvement helps, but you can also tap Esc to pause a crawl mid-run and inspect what has been gathered so far. No need to wait.
The search field accepts more than filenames. Paste a snippet of code and it will narrow the tree to files containing that string. A small thing, but once you know it, you stop browsing manually.
Multiple repositories are not just for switching projects. You can open two side by side, pull context from a library into an application repo, and let the AI work across both. The setup is not obvious on first launch, but it is there in the file menu.
If you run local models, the MCP integration becomes quieter. No API keys. No round trips. The app feels different when latency drops to near zero.
Finally, there is the protocol question. It speaks MCP, the Model Context Protocol. So it plugs into a growing ecosystem of agents without glue code.
