yuuichieguchi/calyx

Summary

Calyx is a native macOS 26 (Tahoe) terminal application built on top of libghostty (the Ghostty terminal engine) with Apple's new Liquid Glass UI aesthetic. It layers significant original features on top of the terminal primitive: tab groups, split panes, a command palette, Git sidebar, AI agent IPC via an embedded MCP server, and a scriptable WKWebView browser with 25 CLI automation commands. It targets macOS 26+ exclusively and requires Xcode 26+.

Great for

people interested in building native macOS terminal tooling, integrating AI agent workflows (Claude Code, Codex) into the terminal via MCP, or exploring the macOS 26 Liquid Glass/Tahoe design system in a real app

Easy wins

• +Write a CONTRIBUTING.md: the repo has zero onboarding docs beyond build steps, no issue templates, and no good-first-issue labels — just documenting the architecture sections would meaningfully lower friction.
• +Fix CI to actually run the test suite: has_ci is true (update-homebrew.yml exists) but there is no workflow that builds or runs CalyxTests/CalyxUITests — all those test files are orphaned from automation.
• +Add universal (arm64 + x86_64) xcframework build: the GhosttyKit.xcframework only contains macos-arm64, making Intel Mac contributors dead on arrival.
• +Add a CHANGELOG or release notes: the repo has a release.sh script but no changelog, making it hard for contributors to understand what changed between versions.

Red flags

• !Only 1 commit in the repo's entire history — this is almost certainly a squash-imported private project, meaning there is no meaningful commit history, no way to understand design decisions, and no prior art for contributors to learn from.
• !The CI pipeline (update-homebrew.yml) only handles Homebrew tap updates — there is no build or test CI, so all 40+ test files are completely unverified by automation.
• !macOS 26 and Xcode 26 are not publicly released as of knowledge cutoff — the barrier to even running this app is an unreleased beta OS, which is an unusual requirement for an open-source project inviting contributors.
• !The GhosttyKit.xcframework ships only macos-arm64 (libghostty-fat.a), excluding Intel Mac developers entirely with no documented path to build x86_64.
• !No CONTRIBUTING.md, no issue templates, no good-first-issues, no help-wanted labels, and 0 open issues despite being public — signals the project may not be actively seeking external contributions despite the MIT license.

Code quality

The project structure is genuinely well-organized — Features/ directory cleanly separates Browser, Git, IPC, Search, Notifications, etc., and GhosttyBridge/ correctly isolates all C FFI calls behind GhosttyFFI.swift. The strict @MainActor enforcement throughout UI/model code is a good Swift 6 practice. However, source_samples were not available for direct code review, and with commit_count of 1 and a single contributor, there's no review history to validate consistency. The test file count is impressive (40+ test files) but has_tests is reported false and no CI runs them, suggesting the tests may exist but not be wired up or passing.

What makes it unique

This is genuinely differentiated — it is not a rehash of iTerm2 or a wrapper around system terminal. The combination of libghostty (Metal-accelerated rendering) + Liquid Glass macOS 26 UI + embedded MCP server for AI agent inter-process communication is novel. The 'AI agents in different panes can message each other via MCP' use case is a real and underexplored problem. The scriptable browser with 25 CLI commands integrated into a terminal is also unusual. The main caveat is that it's essentially a solo project published with a single squash commit, so 'open source' here means 'source available' more than 'community project.'