v0.6.3 — Windows support & cross-platform Whisper installer

What's New

Windows Support

FlexiDesk now runs properly on Windows 10/11. This release fixes multiple Windows-specific issues discovered during testing.

Cross-Platform Whisper Installer

The "Install Whisper" button in the onboarding wizard and Settings now works on all platforms — macOS, Windows, and Linux — without requiring Homebrew.

---

Changes

New Features

• Cross-platform Whisper auto-install — Downloads and installs the correct pre-built whisper-cli binary for your OS and architecture (Windows x64, Linux x64/ARM64). macOS continues to use Homebrew.
• Cross-platform Ollama installer — Windows now downloads and extracts the Ollama binary automatically.

Bug Fixes

• Windows navigation — Switching between sections (Dashboard, Podcast, Caption, etc.) was broken on Windows due to a router incompatibility with the tauri:// protocol. Fixed by switching to HashRouter.
• Whisper install appeared to do nothing — After installation the binary was detected in a subdirectory (where it needs to stay alongside its sibling DLLs), but detection only checked the root. Now searches subdirectories correctly.
• whisper.dll not found — On Windows, the installer was moving only whisper-cli.exe out of its extracted folder, leaving whisper.dll, ggml.dll, and other required DLLs behind. The binary now stays in its extracted folder.
• "Use" button not advancing the wizard — Clicking Use on an already-downloaded Whisper model now correctly activates it and enables the Next button.
• Wrong Whisper download URL — Was using a CUDA build that required an NVIDIA GPU. Now downloads the CPU-only build which works on all machines.

UX Improvements

• CPU incompatibility guidance — If your processor doesn't support AVX2 instructions (Intel 3rd gen / 2012 or older), the installer detects this immediately and shows a clear message with instructions for manually installing a compatible build, instead of silently installing a binary that crashes at transcription time.

---

Platform Support

Platform	Installer	Notes
macOS (Apple Silicon)	.dmg (universal)	Signed & notarized
Windows 10/11 x64	.msi	Requires AVX2 (Intel 4th gen 2013+ / AMD Ryzen)
Linux x64	.AppImage / .deb	—

---

Download

See assets below. macOS users: download the .dmg. Windows users: download the .msi.

v0.6.3-rc6 — build fix

Fix CI build error: replaced missing @tauri-apps/plugin-shell with @tauri-apps/plugin-opener (already installed). AVX2 guidance cards fully functional.

v0.6.3-rc5 — AVX2 compatibility guidance

Pre-release test build

What's new in rc5

AVX2 incompatibility — user-friendly guidance

For CPUs made before 2013 (Intel 3rd gen / Ivy Bridge and older) that don't support AVX2:

• Install-time: after downloading, the installer verifies the binary actually runs. If AVX2 is missing, it cleans up and shows a clear error immediately instead of installing a broken binary silently.
• Onboarding wizard & Settings: instead of a raw error string, users now see a dedicated card that:
  • Explains the issue in plain language ("Your processor (pre-2013) doesn't support AVX2…")
  • Links directly to whisper.cpp releases for a manual/compatible build
  • Explains how to use the Custom binary path field in Settings

Notes

• The prebuilt whisper-bin-x64.zip requires AVX2. Most PCs running Windows 10/11 on modern hardware have AVX2. This only affects users with very old hardware (2012 or earlier).
• For VMs: enable AVX2 CPU passthrough in your hypervisor settings.

Test (on old CPU like i5-3210M)

• Click "Install Whisper" → sees clear AVX2 guidance card, not a raw error
• Card shows link to whisper.cpp releases
• "Skip for now" still works

v0.6.3-rc4 — whisper detection fix

Pre-release test build

Fix

• Install Whisper appeared to do nothing — After a successful install the binary is located in a subdirectory (whisper-bin-x64/) alongside its DLLs. The detection function only checked the managed dir root, so it immediately returned "not found" and the wizard reset. Now detect_whisper_binary() searches root + one level of subdirectories, matching the zip extraction layout.

Combined fixes in this build (rc2 + rc3 + rc4)

• Windows navigation broken (HashRouter)
• Whisper install button silently failing (removed stray installHomebrew() call)
• Wrong download URL (CPU-only zip, no CUDA required)
• whisper.dll not found (binary stays in extracted subfolder with its DLLs)
• Install appears to do nothing (detection now searches subdirectories)

Test checklist (Windows)

• Click "Install Whisper" → progress message appears, completes
• After install wizard advances past "Whisper not found" card
• Model list appears, can select a model
• "Use" button on an already-downloaded model enables Next
• Navigation menu works throughout the app

v0.6.3-rc3 — Windows whisper.dll fix

Pre-release test build — Windows only

Fix

• whisper.dll not found — The zip extraction now keeps whisper-cli.exe in its subfolder alongside whisper.dll, ggml.dll, and other required DLLs. Previously the exe was moved to the parent directory, leaving the DLLs behind and causing a crash on startup.

Test checklist (Windows)

• Install Whisper via the onboarding wizard
• Whisper transcription runs without "whisper.dll was not found" error
• Navigation works (HashRouter fix from rc2 still present)
• "Use" button activates a downloaded model and enables Next

If all pass → promote to v0.6.3 final.

v0.6.3-rc2 — Windows Routing Fix

Pre-release — for testing only.

Fixed in rc2

• Navigation menu now works on Windows (BrowserRouter → HashRouter)
• Whisper Use button now correctly activates model and enables Next
• Whisper install URL fixed (CPU-only, no CUDA required)
• Whisper binary search handles zip subfolders

Please test navigation and Whisper install on Windows.

v0.6.3-rc1 — Windows Whisper Fix (Test Build)

Pre-release — for testing only.

Changes in this build

• Fixed Whisper install on Windows (was failing silently)
• Fixed download URL: now uses CPU-only build (no CUDA required)
• Binary search now handles zip subfolders correctly
• Removed incorrect Homebrew dependency check on Windows/Linux

Please test the Whisper install flow on Windows and report back.

v0.6.2 — Cross-Platform Install, Deck Hub & Anki Export

What's New in v0.6.2

Cross-Platform Support

• Ollama installer now works on macOS, Linux, and Windows
  • macOS: downloads .tgz, extracts via tar
  • Linux: downloads .tar.zst (x86_64 and ARM64), extracts via tar --zstd
  • Windows: downloads .zip, extracts via PowerShell Expand-Archive
  • Windows system Ollama detection via where command and common install paths (C:\Program Files\Ollama)
  • Binary path correctly uses ollama.exe on Windows
• Whisper installer is now fully cross-platform
  • macOS: installs via Homebrew (brew install whisper-cpp) — auto-installs Homebrew if not present
  • Linux & Windows: downloads pre-built binary directly from whisper.cpp GitHub releases
  • No more Homebrew dependency on non-macOS platforms
• Onboarding UI no longer mentions Homebrew — shows a neutral "Install automatically" message on all platforms

Deck Hub

• New Deck Hub for creating and managing flashcard decks
• Text to Deck: generate flashcards from any text using AI
• Image to Deck: scan an image and extract vocabulary into a deck
• Add to Deck sheet: quickly save items from podcast or reading sessions to a deck
• Deck detail page improvements: browse and manage cards within a deck

Anki Export

• Export any flashcard deck to Anki .apkg format
• Supports front/back cards with optional example sentences
• Accessible from the Deck Hub export panel

Podcast

• Cloud sync dialog for syncing podcast subscriptions
• Additional podcast backend commands

Bug Fixes

• Fixed hardcoded Homebrew paths causing install errors on Linux and Windows
• Fixed Ollama binary name on Windows (.exe extension)
• Fixed Whisper detection on Windows using where command

v0.6.1 — Caption Module, macOS Build, README Updates

✨ What's New

Features

• Caption module enabled — Live audio capture with real-time Whisper transcription is now available
• Tutor & Caption modules activated in feature flags
• Improved Whisper setup — Better error handling for sidecar and transcription

CI/CD

• macOS build restored — Release workflow now builds for Apple Silicon (aarch64) without notarization
• Users can install via right-click → Open to bypass Gatekeeper

Documentation

• Added per-module versioning to README (stable / beta / planned)
• Updated Dashboard description to reflect actual features
• Marked Caption and AI Tutor as available modules

v0.6.0 — AI Tutor Redesign

✨ AI Tutor Redesign

Major overhaul of the AI Tutor experience with a new conversation-first flow and dramatically expanded content.

Features

• Conversation History — Landing page shows all past conversations with mode, language, CEFR level, and message count
• Conversation Detail View — Click any conversation to see scores (grammar, vocabulary, fluency), corrections, new words learned, and the full chat transcript
• 3-Step Setup Wizard — New conversation creation flows through: Practice Mode → Language & Model → Topic/Scenario
• 63 Role-Play Scenarios — Expanded from 10 to 63 scenarios across 8 categories (Daily Life, Travel, Social, Professional, Academic, Emergency, Creative, Health)
• Searchable Scenario List — Filter scenarios by name, description, or category with real-time search
• Custom Scenarios — Write your own role-play situation
• Suggested Topics — 20 clickable topic chips for free conversation mode
• Deck Selector — Dropdown of your flashcard decks replaces manual ID input
• Ollama Model Selector — Shows downloaded models when Ollama is connected

Improvements

• AI responses are shorter and more interactive — always ends with a follow-up question
• AI encourages the student to speak more instead of lecturing
• Natural conversation flow instead of forced topic transitions
• Resume active conversations from the detail view