Skip to content

Releases: alexbic/hush

HUSH v1.1 — Smart Grid Layout

Choose a tag to compare

@alexbic alexbic released this 25 Jun 05:21

What's new in v1.1

Panel grid system — rewritten from scratch

Reset ([🔄]) now works correctly at any screen size.

The layout pattern is chosen automatically based on available screen space:

  • 3×3 — classic cross (cfg above, providers left, history below, editor right)
  • 3×2 — providers left, main center, cfg right; history below main, editor below providers (most MacBooks)
  • 2×2 — compact 2×2 grid

All windows are placed as a group centered on screen — the bounding box of all 5 windows is calculated first, then the entire group is shifted to fit. No overlaps, no windows hidden behind screen edges.

New placement rule: a diagonal cell is only used when at least one face-adjacent neighbor is already occupied. Windows never connect corner-to-corner with empty space between them.

Live repositioning fixed: when dragging the main window toward a screen edge, panels that go off-screen now flip to the opposite side. Previously the algorithm would attempt to build a 4th horizontal column that went off-screen.

UI consistency

  • All buttons unified to [UPPERCASE] style with square brackets across all three languages (RU/EN/ES).
  • The [SCENARIO] / [СЦЕНАРИЙ] button color is now theme-aware: C_YEL is defined per theme and updates live on theme switch. On light themes (paper, sky, sand, arctic) it switches to dark amber so text remains readable.

Provider monitoring

  • Scenario buttons turn red when the assigned LLM provider is unreachable.
  • Status is polled every 30 s via background probe threads.
  • The scenario list refreshes automatically on each status change (no restart needed).

Full changelog: see README → Changelog

HUSH v1.0

Choose a tag to compare

@alexbic alexbic released this 22 Jun 20:43

HUSH v1.0 — initial public release

Voice-to-text with LLM post-processing for macOS.

Features

  • Silent mode (Right ⌥): hold to record, release to transcribe, 4-second LLM countdown
  • Full mode (⇧⌥): multi-block dictation with scenario selection and review
  • Expanded mode: double-click title bar to resize the window
  • Scenario editor: custom LLM system prompts, model override per scenario
  • History panel: last 50 transcriptions with tabs (All / Sessions / Blocks)
  • LLM providers: Ollama (local), Anthropic, OpenAI, GLM — with model picker dropdowns
  • 8 color themes switchable in settings
  • Language switcher: Russian / English / Spanish
  • Cluster mode: 2×2 panel layout with magnet drag
  • On-device ASR: Parakeet TDT 0.6B via CoreML / Apple Neural Engine — private, no cloud mic
  • Text injection via Accessibility API — no clipboard, no flicker

Requirements

  • macOS 13+ (Ventura or newer)
  • Python 3.14 (brew install python@3.14)
  • pip3.14 install pyobjc sounddevice pynput anthropic openai

Quick start

git clone https://github.com/alexbic/hush.git && cd hush
pip3.14 install pyobjc sounddevice pynput anthropic openai
bash build_app.sh && open HUSH.app