Larkyn 0.2.3

New

• About page — what Larkyn is, the technology inside it (Whisper, Gemma 4 via Ollama, PySide6/Fluent, SQLite), and credits, with a link back to this repo.
• Start with Windows — new switch in Settings → Startup. Larkyn registers itself to launch minimized to the tray at sign-in (per-user, no admin), so dictation is always one hotkey away. Works alongside or instead of the installer's startup option — no double-launching.

Install / upgrade

Upgrading from 0.2.x: just run LarkynSetup.exe — installs in place, settings and history kept.

First time? Follow the 4-step setup in the v0.2.2 release notes or the README: install Ollama, ollama pull gemma4:e2b-it-qat, run LarkynSetup.exe, dictate.

Verify your download against SHA256SUMS.txt. Unsigned installer: if SmartScreen warns, More info → Run anyway.

Larkyn 0.2.2

Performance and clarity update.

Improved

• First dictation is now fast. The speech model used to load on first use, costing 5–8 seconds on the first dictation of every session. Larkyn now warms it up in the background at launch — first-dictation transcription dropped to under one second in testing.
• Raw mode explained in the UI. The Home screen now notes that Raw mode pastes exactly what you said, skipping the AI cleanup.

---

🚀 First-time setup (about 10 minutes, mostly downloads)

Larkyn needs one companion program: Ollama, which runs the AI rewriting model locally on your PC. Everything stays on your machine — nothing goes to the cloud.

Step 1 — Install Ollama

1. Go to https://ollama.com/download and click Download for Windows
2. Run OllamaSetup.exe. When it finishes, Ollama runs quietly in the background (a llama icon appears in your system tray). Nothing else to configure.

Step 2 — Download the AI model (one time, ~4.3 GB)

1. Open PowerShell: press the Windows key, type powershell, press Enter
2. Copy-paste this and press Enter:

   ollama pull gemma4:e2b-it-qat
   

3. Wait for the download to finish, then confirm it's installed:

   ollama list
   

   You should see gemma4:e2b-it-qat in the list. ✅

Step 3 — Install Larkyn

1. Download LarkynSetup.exe from the Assets section below
2. Run it. If Windows SmartScreen warns (the installer isn't code-signed yet), click More info → Run anyway
3. Follow the wizard — no admin rights needed. Optional: desktop shortcut and start-at-sign-in.

Step 4 — Dictate

1. Launch Larkyn from the Start menu. Your very first dictation downloads the speech-recognition model (~1.6 GB, one time) — give it a few minutes; after that it's instant.
2. Click into any text box (email, document, chat…), hold Ctrl+Alt+Space, speak naturally, release. Polished text appears where your cursor is.
3. Make it yours: Settings → Global hotkey to pick a different combo (e.g. Ctrl+Win), Settings → AI model → Test connection to verify your setup, and the Home screen to switch writing profiles.

---

Upgrading from 0.2.x

Just run the new LarkynSetup.exe — it installs in place; your settings and history are kept. (Already did steps 1–2? Skip them.)

Verify your download (optional)

Get-FileHash .\LarkynSetup.exe -Algorithm SHA256

The hash must match the value in SHA256SUMS.txt.

Something not working?

Larkyn tells you what's wrong in plain English (e.g. "Ollama doesn't appear to be running" or "the model isn't installed") — and Settings → AI model → Test connection diagnoses the AI setup in one click.

Larkyn 0.2.1

Quality-of-life update based on first user feedback.

Fixed

• Clear error messages when the AI backend isn't ready. Previously, dictating before pulling the model showed a cryptic Client error '404 Not Found' for url 'http://localhost:11434/api/chat'. Larkyn now tells you exactly what to do:
  • Model not installed → "The AI model 'gemma4:e2b-it-qat' isn't installed in Ollama. Open a terminal and run: ollama pull gemma4:e2b-it-qat"
  • Ollama not running → "Ollama doesn't appear to be running. Start Ollama, then dictate again."
  • Timeouts and other backend errors now show concise, actionable text.

Install

1. Install Ollama (v0.30.7+) and pull the model: ollama pull gemma4:e2b-it-qat
2. Run LarkynSetup.exe (no admin required) — upgrading from 0.2.0 installs in place, settings and history are kept
3. First dictation downloads the Whisper speech model (~1.6 GB, one time)

Verify your download against SHA256SUMS.txt. The installer is not code-signed yet; if SmartScreen warns, choose More info → Run anyway.

Larkyn 0.2.0

First public release of Larkyn — a privacy-first, system-wide voice-to-writing assistant for Windows 11.

Hold a hotkey, speak naturally, release: your speech becomes polished written text, pasted straight into whatever app you're using. 100% local — no cloud, no telemetry.

Install

1. Install Ollama (v0.30.7+) and pull the model: ollama pull gemma4:e2b-it-qat
2. Run LarkynSetup.exe (no admin required)
3. First dictation downloads the Whisper speech model (~1.6 GB, one time)

Highlights

• Push-to-talk global hotkey (hold to record, release to send) or toggle mode — any combo, including modifier-only like Ctrl+Win
• Local GPU Whisper transcription (CUDA, with CPU fallback) + local LLM rewriting via Ollama
• Windows 11 Fluent UI: dashboard, searchable history, writing profiles, personal vocabulary, live-apply settings
• 7 built-in writing profiles + custom profiles; smart spoken commands

Notes

• Verify your download against SHA256SUMS.txt
• The installer is not code-signed yet; Windows SmartScreen may warn — choose More info → Run anyway
• Requires Windows 11 x64, 16 GB RAM; NVIDIA GPU recommended