PMT

Version: v0.0.91

中文说明

PMT is a lightweight native macOS app for rewriting selected text and preview dictation into structured prompts with global shortcuts.

Core Workflows

1. Select text in any editable field.
2. Press the rewrite hotkey.
3. PMT copies the selection, rewrites it through the configured model provider, and pastes the result back in place.

Preview dictation workflow:

1. Press the dictation hotkey to start recording.
2. Press the same hotkey again to stop recording.
3. PMT transcribes locally with WhisperKit, sends the transcript once to the configured remote model for structured rewriting, and pastes the result at the cursor.

Short dictation snippets are inserted directly without model rewriting.

Features

• Native macOS app built with Swift, SwiftUI, and AppKit.
• Global selected-text rewrite workflow.
• Default rewrite hotkey: Ctrl + X.
• Configurable custom OpenAI-compatible endpoint.
• GitHub OAuth authentication for GitHub Copilot access.
• Full model list loading, model selection, and model latency testing.
• Built-in prompt styles plus custom prompt support.
• Preview dictation feature for Apple Silicon Macs.
• Local WhisperKit transcription with Base and Small model options.
• Apple Silicon Metal acceleration for dictation transcription.
• Two-step dictation flow: local speech transcription, then remote model structured rewriting.
• Stage indicators for dictation and model processing.
• Start and stop dictation sounds.
• Optional floating 🤔 icon: click to rewrite the current text field, right-click to open the settings panel, and long-press to drag.
• The floating icon is enabled once during this version migration and can then be disabled and saved from the panel.
• Local action log for hotkeys, permissions, model requests, and rewrite results.
• Optional local advice report at ~/Documents/PMT/advice.md, generated from PMT rewrite and dictation history.
• Optional Telegram Bot delivery for the final advice summary.
• Chinese and English interface switching.
• In-app update checking and installation through Sparkle.

Project Structure

• Sources/PMT: application source code.
• Resources: app icon assets.
• scripts/build-app.sh: builds dist/PMT.app.
• scripts/package-release.sh: builds release ZIP, DMG, and Sparkle appcast.
• dist/PMT.app: generated app bundle after packaging.

Run

swift run PMT

Build

./scripts/build-app.sh
open dist/PMT.app

Release Updates

PMT uses Sparkle for in-app updates. The app reads its update feed from:

https://raw.githubusercontent.com/ink1ing/pmt/main/appcast.xml

To build a Sparkle update ZIP, a user-facing DMG, and regenerate the Sparkle appcast:

scripts/package-release.sh 0.0.91 91

Upload both files to the matching GitHub release tag:

• release/appcast/PMT-0.0.91.zip: used by Sparkle in-app updates.
• release/downloads/PMT-0.0.91.dmg: user-facing installer download.

Then commit and push the generated appcast.xml.

Permissions

PMT needs these macOS permissions for the full workflow:

• Accessibility: activate the target app and send copy or paste shortcuts.
• Input Monitoring: receive global hotkeys while another app is focused.
• Microphone: record dictation audio when preview dictation is enabled.

The settings panel includes checks and request actions for these permissions.

Privacy

When advice generation is enabled, PMT records only the text handled inside PMT rewrite or dictation workflows. This history is stored locally at:

~/Library/Application Support/PMT/advice-history.jsonl

Advice generation is disabled by default. PMT does not create advice history while it is disabled. When enabled, PMT reads local history and sends the minimum necessary recent snippets to the model provider configured by the user, then writes a short summary to:

~/Documents/PMT/advice.md

Telegram is optional and only sends the final advice summary, not the raw local history.

API Compatibility

PMT supports two model channels:

• OpenAI-compatible endpoint: model list and chat completions.
• GitHub OAuth: GitHub Copilot device authorization, full model loading, and chat completions.

For custom endpoints, PMT uses:

• GET {endpoint}/models
• POST {endpoint}/chat/completions

Default endpoint:

https://api.openai.com/v1

Custom gateways and local providers are supported as long as they expose compatible endpoints.

Status

v0.0.91 improves configuration upgrade resilience, removes dead Keychain code, adds an explicit Quit button, and removes hard-coded app version usage from GitHub requests.