Independent macOS App Suite

Hikari Brandan built independent desktop utilities and experiments for his own workflow, working on local processing, native interfaces, shortcuts, permissions and debugging.

Some of the problems I build for are my own: repeating a typing task, untangling files, watching system activity or making another device useful.

• Keep each application centered on one practical task.
• Use native macOS interfaces for the Swift utilities and a Tauri / React shell for Screen Bridge.
• Prefer local processing where the product supports it, with explicit boundaries for optional cloud use.

Personal use is a useful feedback loop. It exposes processing delays, unreliable automation, setup friction and compatibility issues that a polished screenshot cannot show.

These applications have different levels of maturity. Screen Bridge is experimental, and several current binaries are staged for smoke testing.

The app notes below distinguish public source, verified release downloads and known limitations.

Some hotkeys were not recognized consistently, and microphone permission could be requested again after relaunch. I tested different shortcuts and repeated macOS launch and permission scenarios to improve those flows.

• Swift / SwiftUI
• AppKit
• React / Tauri
• Rust
• macOS APIs
• Local processing

A global keyboard shortcut records speech, transcribes it locally and inserts text into the active Mac application. Optional local cleanup removes filler words and improves formatting.

macOS 14+ · Apple Silicon

Product decisions

• Keep transcription local with WhisperKit and Core ML.
• Make dictation fit into existing applications through a global shortcut.
• Add vocabulary, snippets, language settings and transcription history for repeat use.

What I learned

The useful part is the complete capture-to-insertion workflow. Model setup, processing time and daily reliability matter as much as the transcription itself.

Limitations

• Speech models must be downloaded first; longer recordings require processing time.
• The published README describes lag developing during daily use. Its cause is not yet diagnosed.
• The v1.1.0 Apple Silicon build is ad-hoc signed and is not notarized.

A personal cleanup and organization utility with Smart Sanitize scans, reviewable candidates, Auto-Flow filing rules, SHA-256 duplicate checks and local activity.

macOS 14+ · Local processing

Product decisions

• Use local rules and file scanning; no backend, account or API key.
• Keep cleanup candidates reviewable before acting.
• Use SHA-256 to compare file content when checking duplicates.

What I learned

File automation needs dependable controls. A reviewable scan is more useful than an ambitious cleanup promise that cannot be trusted.

Limitations

• Manual scanning and review are the most dependable path.
• Automatic filing and deletion are still maturing and have not been consistently reliable.
• The latest ZIP is staged for smoke testing; the public source is the available build path.

An experimental Mac-to-device display project. It creates a virtual Mac display and streams it to a compatible browser on the local network, with QR pairing and browser input relay.

Experimental · Trusted local network

Product decisions

• Use a browser as the receiving surface to reduce device setup.
• Connect a Tauri desktop shell with Rust networking and macOS graphics APIs.
• Explore virtual display creation through an Objective-C bridge and CGVirtualDisplay.

What I learned

Getting pixels to another device is only the beginning. Latency, synchronization, compatibility and unsupported system APIs determine whether an experiment is ready for daily use.

Limitations

• Experimental: streams can lag and compatibility varies.
• CGVirtualDisplay is a private, unsupported Apple API.
• The local connection is unencrypted and is intended for trusted networks.
• Audio capture requires macOS 14.2+. Current 0.1.1 source is staged for creator testing.

A lightweight system dashboard for CPU, memory, network, disk and top processes, with short rolling charts. It runs locally without an external backend.

Local metrics · No external backend

Product decisions

• Use Mach and BSD system APIs for local metrics.
• Keep the interface compact and glanceable.
• Show recent trends with Swift Charts instead of building a complex monitoring suite.

What I learned

A small utility benefits from a clear boundary. A useful overview does not need to become a replacement for every system diagnostic tool.

Limitations

• Memory categories are estimates; charts retain 60 recent samples.
• It is a compact overview, not an Activity Monitor replacement.
• Temperature, fan and battery monitoring are not included.
• The 1.0.1 Apple Silicon binary is staged for smoke testing.

A native AI chat app with Ollama model selection, streaming responses, conversation search and history, custom instructions, and optional OpenAI or Anthropic providers.

macOS 14+ · Ollama installed separately

Product decisions

• Connect to the user’s local Ollama service rather than bundling a model runtime.
• Keep conversations available through local history and search.
• Store optional provider keys in the macOS Keychain.

What I learned

Local and cloud modes need explicit boundaries. Setup, model availability and which context leaves the machine are part of the product experience.

Limitations

• Ollama and its models are separate installations and are not bundled.
• Cloud mode requires the user’s provider API key and can incur provider charges.
• Cloud mode sends relevant conversation history and system context to the selected provider.
• The latest 1.0.1 binary is staged for smoke testing; public source is available.