DictaPulse 0.1.1

DictaPulse v0.1.1

A packaging-focused patch release. No app behavior changes from v0.1.0 — this makes
DictaPulse cleanly installable as a download-and-run AppImage and unblocks distro
packaging (AUR).

What's changed

• Fix: the .desktop install now honors DESTDIR, so staged/packaged installs
  (AppImage, AUR package()) build correctly instead of trying to write to /usr.
• New: prebuilt Vulkan AppImage attached below, verified on KDE Plasma 6 Wayland.

Install

• Download & run (AppImage, Vulkan): grab DictaPulse-0.1.1-x86_64.AppImage below, then:

  chmod +x DictaPulse-0.1.1-x86_64.AppImage
  ./DictaPulse-0.1.1-x86_64.AppImage

  Needs KDE Plasma 6 (Wayland), a running ydotoold, and host ffmpeg (present on most desktops). GPU use needs a Vulkan driver; otherwise it falls back to CPU.
• Arch / CachyOS (build from source):

  git clone https://github.com/Silverhairfx/DictaPulse.git
  cd DictaPulse && ./scripts/install.sh      # add --vulkan / --cuda / --rocm for GPU

• See the README for full requirements, first-run setup, and the model picker.

Full changelog: v0.1.0...v0.1.1

DictaPulse 0.1.0

DictaPulse v0.1.0 — first public beta 🎙️

Local AI voice dictation for KDE Plasma 6 (Wayland): press a shortcut, speak, and polished text is typed straight into whatever app you're focused on. Fully offline — audio never leaves your machine.

This is a beta. It's daily-driver-usable, but expect rough edges as features stabilize. Bug reports and ideas during beta directly shape what ships — Issues for bugs, Discussions for features.

Highlights

• 🎙️ Global shortcut → dictate anywhere — works in any focused window via KGlobalAccel, even from the tray. Toggle, push-to-talk, or auto-stop-on-silence.
• 🧠 Local Whisper transcription — whisper.cpp runs entirely on-device. CPU out of the box; optional Vulkan / CUDA / ROCm GPU builds.
• 🌊 Animated listening overlay — a floating pill with a live waveform; position, size, opacity, sounds, and reduced-motion all configurable.
• ⚡ Direct text insertion — typed into the active app via wtype / ydotool, with clipboard as a fallback.
• ✨ Transcript cleanup, your way — offline rules engine, a local LLM (Ollama / LM Studio), or a remote API (Anthropic / OpenAI / any OpenAI-compatible endpoint). API keys live in your system keyring.
• 👤 AI Profile that learns you — Refine, Personalize (dictionary, templates, usage dashboard), and Devs Voice, all fed to whichever AI you've connected.
• 🎯 Per-app output rules + 🎛️ built-in model manager + 🌐 99 languages with constrained auto-detect.
• 🎨 Tactile "clay" design system with matching dark & light themes.
• 🛡️ Privacy-first — no telemetry, no recordings on disk by default, cloud cleanup strictly opt-in.

Install

• Download & run (AppImage, Vulkan): grab DictaPulse-0.1.0-x86_64.AppImage below, then:

  chmod +x DictaPulse-0.1.0-x86_64.AppImage
  ./DictaPulse-0.1.0-x86_64.AppImage

  Needs KDE Plasma 6 (Wayland), a running ydotoold, and host ffmpeg (present on most desktops). GPU use needs a Vulkan driver; otherwise it falls back to CPU.
• Arch / CachyOS (build from source):

  git clone https://github.com/Silverhairfx/DictaPulse.git
  cd DictaPulse
  ./scripts/install.sh          # add --vulkan / --cuda / --rocm for GPU

• See the README for full requirements, first-run setup, and the model picker.

Requirements

• KDE Plasma 6 on Wayland, Qt 6.9+, KDE Frameworks 6, a C++20 compiler.
• A running ydotoold (or working wtype) for text injection.
• A Whisper model is downloaded on first run from the in-app model manager.

Known limitations

• KDE Plasma 6 / Wayland only for now; the core is structured for future Windows / macOS / mobile ports.
• Beta — see open Issues for current rough edges.