Thalia: Plan, approve and review your Muse Code agent on Mac

Thalia is a free native Mac app for Meta's Muse Code CLI. Plan mode restarts Muse read-only until you approve the plan. Every command is shown in full, step by step, before Allow. Review lets you run your checks and approve the exact snapshot, and any later edit invalidates it. Checkpoints, MCP connectors, voice mode and an embedded terminal are built in. Uses your own Muse access, sends no telemetry, and runs on Apple Silicon Macs with macOS 27.

Hi Product Hunt 👋 I'm Jared, and I built Thalia, a free native Mac app for Meta's Muse Code CLI. I wanted the speed of letting an agent work without losing track of what it was about to do or what it actually changed. Thalia is built around that loop: • Plan mode that's enforced, not just a prompt. Muse restarts with writes and shell disabled, and Thalia only shows Plan once that's true. Approve the plan and it builds. • See the whole command before Allow. Compound commands show each step ("Step 2 of 3"), and long commands stay locked until you expand them. • Approve an exact snapshot. Run your checks beside the diff; any later edit invalidates the approval. • Checkpoints before each turn, with a preview of what a restore will change. • MCP connectors with sign-in and verification, voice mode, routines and an embedded terminal. It's native SwiftUI, not a web wrapper, it sends us no analytics or telemetry, and it uses your own Muse access. It needs an Apple Silicon Mac on macOS 27. Install from usethalia.com or with: brew install --cask mabry-ventures/tap/thalia I'd love to hear how you use coding agents today and what you'd want Thalia to do next.

I like the idea of seeing every command before it actually runs. That extra step could make working with coding agent feel a lot less risky.

@jiyo_root Thanks, jiyo! That was the starting point for Thalia. Allow stays locked until the whole command is on screen, and compound commands are shown one step at a time, so you never approve something you haven't read. If you try it with Muse Code, I'd love to hear where the approvals feel too slow or too chatty.

I like the idea of seeing every command before it actually runs. That extra step could make working with coding agent feel a lot less risky.