- transcribe: catch model load failures on CUDA and retry with CPU
- hardware detect: test CUDA runtime actually works (torch.zeros on cuda)
before recommending GPU, since CPU-only builds detect CUDA via driver
but lack cublas/cuDNN libraries
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Exclude ctranslate2.converters from PyInstaller bundle — these modules
import torch at module level causing circular import crashes, and are
only needed for model conversion (never used at runtime)
- Defer all heavy ML imports to first handler call instead of startup,
so the sidecar can send its ready message without loading torch/whisper
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Fix is_running() to check actual process liveness via try_wait()
instead of just checking if the handle exists
- Auto-restart sidecar on pipe errors (broken pipe, closed stdout)
with one retry attempt
- Hide sidecar console window on Windows (CREATE_NO_WINDOW flag)
- Log sidecar stderr to sidecar.log file for crash diagnostics
- Include exit status in error message when sidecar fails to start
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Pushes from within a workflow don't trigger other workflows in Gitea,
so the separate tag-triggered build files never ran. Moved all 3
platform build jobs into release.yml with needs: bump-version so they
run directly after the version bump, tag, and release creation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- New release.yml: bumps patch version, commits with skip-ci marker, tags, creates Gitea release
- Build workflows now trigger on v* tags only (not branch push)
- Simplified upload steps: use tag directly, retry loop for release lookup
- Fix macOS: install jq if missing
- Sync python/pyproject.toml version to 0.2.0
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Pushing to main + a tag triggered 6 workflows (3 per trigger).
Now only main pushes trigger builds. The upload step detects version
tags on the current commit via git tag --points-at HEAD.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
AppImage bundler compresses the entire sidecar.zip into squashfs,
causing builds to hang/timeout. Limit targets to deb (Linux),
nsis+msi (Windows), and dmg (macOS).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Invoke-RestMethod loads entire files into memory, causing connection
failures on 360MB+ installer files. Switch to curl which streams
the upload.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Upload step now runs on both main pushes and v* tag pushes
- Tag pushes create a versioned release (e.g., "Voice to Notes v0.2.0")
- Main pushes update the "latest" prerelease as before
- Windows: filter for *-setup.exe to avoid uploading non-installer binaries
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The TAURI_CONFIG env var approach for resources wasn't being applied
by the NSIS bundler, so sidecar.zip was never included in the installer.
- Add resources: ["sidecar.zip"] directly to tauri.conf.json
- build.rs creates a minimal placeholder zip for dev builds so
compilation succeeds even without the real sidecar
- Remove TAURI_CONFIG env var from all CI workflows (no longer needed)
- Add sidecar.zip to .gitignore (generated by CI, not tracked)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Tauri's build script overflows the stack when processing resource globs
matching thousands of files from PyInstaller's ML output (torch, pyannote).
Instead of bundling the sidecar directory directly:
- CI zips the sidecar output into a single sidecar.zip
- Tauri bundles just the one zip file (no recursion)
- On first launch, Rust extracts the zip to the app data directory
- Versioned extraction dir (sidecar-{version}) ensures updates re-extract
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Tauri's externalBin only bundled the single sidecar executable, but
PyInstaller's onedir output requires companion DLLs and _internal/.
The binary was also renamed with a target triple suffix that
resolve_sidecar_path() didn't look for, causing it to fall back to
dev mode which used a compile-time CI path (CARGO_MANIFEST_DIR).
- Switch from externalBin to bundle.resources to include all sidecar files
- Pass Tauri resource_dir to sidecar manager for platform-aware path resolution
- Remove rename_binary() since externalBin target triple naming is no longer needed
- Remove broken production-to-dev fallback that could never work on user machines
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Removes the artifact upload/download overhead between sidecar and app
build steps. Each platform now runs as a single job: build sidecar,
copy it into src-tauri/binaries, build Tauri app, upload to release.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Use [System.Uri]::EscapeDataString for proper encoding of filenames
containing spaces in the Gitea API URL. Add size logging and error handling.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Use curl -T (streaming) instead of --data-binary (loads into memory)
to handle large .deb/.AppImage files
- URL-encode spaces in filenames for the Gitea API
- Use IFS= read -r to handle filenames with spaces
- Add HTTP status code logging for upload debugging
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Each platform (Linux, macOS, Windows) now has its own workflow file
that builds the sidecar, builds the Tauri app, and uploads to a shared
"latest" release independently. A failure on one platform no longer
blocks releases for the others.
- build-linux.yml: bash throughout, apt for deps
- build-macos.yml: bash throughout, brew for deps
- build-windows.yml: powershell throughout, choco for deps
- All use uv for Python, upload to shared "latest" release tag
- Each platform replaces its own artifacts on the release
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Linux: add xdg-utils to system deps (provides xdg-open needed by
Tauri's AppImage bundler)
- Windows: replace dtolnay/rust-toolchain action (uses bash internally)
with direct rustup install via PowerShell
- Unix: install Rust via rustup.rs shell script instead of GitHub action
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
pyannote.audio requires ffmpeg at import time (torchcodec loads
FFmpeg shared libraries). Install via brew (macOS), apt (Linux),
choco (Windows) before building the sidecar.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Windows runner doesn't have bash. Split Python setup and build steps
into Unix (default shell) and Windows (powershell) variants.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- uv pip --python works better with the venv directory path than the
python binary path (avoids "No virtual environment found" on Windows)
- Add .exe suffix to Windows python path for non-uv fallback
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- build_sidecar.py: pip_install() now includes 'install' in the command,
callers pass only package names (was doubling up as 'uv pip install install torch')
- CI: set shell: bash on uv steps so Windows doesn't try to use cmd.exe
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Unix runners use the bash install script, Windows uses the PowerShell
installer. Both check if uv is already present first.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
astral-sh/setup-uv is not available on Gitea's action registry.
Use the official install script instead, skipping if uv is already present.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- CI: install uv via astral-sh/setup-uv, use uv to install Python
and run the build script (replaces setup-python which fails on
self-hosted macOS runners)
- build_sidecar.py: auto-detects uv and uses it for venv creation
and package installation (much faster), falls back to standard
venv + pip when uv is not available
- Remove .github/workflows duplicate
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Set AGENT_TOOLSDIRECTORY via step-level env on setup-python (not
GITHUB_ENV which only applies to subsequent steps)
- Use runner.temp for toolcache dir (always writable, no sudo needed)
- Remove .github/workflows/build.yml to prevent duplicate CI runs
- Remove unused Windows env check step
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Set AGENT_TOOLSDIRECTORY to a workspace-local path so setup-python
doesn't need /Users/runner or sudo access.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The app name is already in the window title bar, so the in-header
"Voice to Notes" heading was redundant and had poor contrast.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When editing a segment, word timing is now intelligently redistributed:
- Spelling fixes (same word count): each word keeps its original timing
- Word splits (e.g. "gonna" → "going to"): original word's time range
is divided proportionally across the new words
- Inserted words: timing interpolated from neighboring words
- Deleted words: remaining words keep their timing, gaps collapse
This preserves click-to-seek accuracy for common edits like fixing
misheard words or splitting concatenated words.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When the edited text has the same word count as the original (e.g. fixing
"Whisper" to "wisper"), each word keeps its original start/end timestamps.
Only falls back to segment-level timing when words are added or removed.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The display renders segment.words (not segment.text), so editing the text
field alone had no visible effect. Now finishEditing() rebuilds the words
array from the edited text so the change is immediately visible.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Project files (.vtn):
- Save Project: serializes transcript, speakers, audio path to JSON file
- Open Project: loads .vtn file, restores audio/transcript/speakers
- User chooses filename and location via save dialog
- Replaces SQLite-based project persistence (DB commands remain for future use)
- Text edits update in-memory store immediately, persist on explicit save
- Fix Windows path separator in project name extraction
AI chat:
- Markdown rendering in assistant messages (headers, lists, bold, code)
- Better visual distinction with border-left accents
- Styled markdown elements for dark theme
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add update_segment Tauri command (calls existing update_segment_text query)
- Wire onTextEdit handler from TranscriptEditor to invoke update_segment
- Edits are saved to SQLite immediately when user presses Enter
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Project persistence:
- save_project_transcript command: persists segments, speakers, words to SQLite
- load_project_transcript command: loads full transcript with nested words
- delete_project command: soft-delete projects
- Auto-save after pipeline completes (named from filename)
- Project dropdown in header to switch between saved transcripts
- Projects load audio, segments, and speakers from database
AI chat improvements:
- Markdown rendering in assistant messages (headers, lists, bold, italic, code)
- Better message spacing and visual distinction (border-left accents)
- Styled markdown elements matching dark theme
- Improved empty state and quick action button sizing
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
upload-artifact@v4 and download-artifact@v4 require GitHub's backend
and are not supported on Gitea. v3 works with Gitea Actions.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Create /Users/runner directory on macOS before setup-python (permission fix)
- Use `python -m pip` everywhere instead of calling pip directly (Windows fix)
- Refactor build_sidecar.py to use pip_install() helper via python -m pip
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
