SpotAI v1.4.4

SpotAI v1.4.4

Changes since v1.4.3:

✨ New Features

• v1.4.4 - RAG doc selection, thinking models, direct send, stop/retry and context CTA

📚 Documentation

• add CHANGELOG entry for v1.4.3

SpotAI v1.4.3

SpotAI v1.4.3

Changes since v1.4.2:

📚 Documentation

• add CHANGELOG entry for v1.4.2

🧹 Maintenance

• fix test-results/ ignore pattern (was broken with markdown backticks)

📦 Other

• feat(rag,ai): v1.4.3 - AI-suggested document actions with one-click ask and regenerate

SpotAI v1.4.2

SpotAI v1.4.2

Changes since v1.4.1:

📚 Documentation

• translate remaining README feature bullets to English
• add CHANGELOG entry for v1.4.1

🧹 Maintenance

• untrack build artifacts (dist/) committed by the merged PR

📦 Other

• feat(rag,ui): v1.4.2 - RAG auto-context toggle, context badge, clickable citations and reveal in folder
• feat(rag,stream): RAG doc sync, suggested questions and streaming token batching

SpotAI v1.4.1

SpotAI v1.4.1

Changes since v1.4.0:

📚 Documentation

• add CHANGELOG entry for v1.4.0

📦 Other

• feat(rag,drop): v1.4.1 - native file drag & drop fix + many more RAG formats

SpotAI v1.4.0

SpotAI v1.4.0

Changes since v1.3.4:

📚 Documentation

• Release v1.4.0 with RAG Local, Real Embeddings, CLI Injection & Multilingual Voice
• add CHANGELOG entry for v1.3.4

📦 Other

• feat(rag,cli): v1.4.0 - local RAG (sqlite-vec + Ollama embeddings), /exec CLI injection, advanced LLM sampling params

SpotAI v1.3.4

SpotAI v1.3.4

✨ New Features

• 🎚️ Recognition Language Setting: Whisper now transcribes in the language you pin (Auto / English / Español / Deutsch / Português / Français) instead of relying on its unreliable auto-detection, which on the tiny model tends to assume English and produce nonsense on other languages. When nothing is pinned, the app defaults to your interface language automatically — a Spanish UI user gets Spanish dictation without touching Settings.
• 📦 Whisper Model Size Selector: Choose Tiny (~75 MB), Base (~145 MB) or Small (~466 MB) right in the download panel. Bigger models are noticeably more accurate on non-English speech. Switching only downloads the model you pick — the binary is shared and already-downloaded sizes switch instantly.

🧪 Tests

• 🧩 New Rust tests cover the model-id resolution and the language→-l flag mapping; new E2E tests verify the engine/mic/language startup sync and the model-switch flow (an uninstalled model shows the download button, a downloaded one shows ready instantly).

📚 Documentation

• 📖 The README Voice section now documents the model picker and the recognition language (tip: pick Base/Small + Español for Spanish dictation), with screenshots of the Base panel and the transcribed result.

SpotAI v1.3.3

SpotAI v1.3.3

🐛 Bug Fixes

• 🎙️ Dictation Actually Works: Three fixes make transcription reliable end-to-end. The recorded WAV header now carries the real device sample rate (no more hardcoded 16 kHz that made Whisper hear desktop mics at 44.1/48 kHz ~3× slower and fail), the selected voice engine + microphone are synced to the backend at startup (no more reverting to the broken native engine after a restart), and silent captures show an honest "No speech detected" instead of injecting a [BLANK_AUDIO] token.

🧪 Tests

• 📼 WAV Header Tests: New Rust tests prove write_wav writes the real sample rate (48 kHz ramp + a 5-rate sweep from 8 to 48 kHz) — regression-proof against the hardcoded header.
• 🔁 Startup Sync E2E: Playwright tests verify the selected Whisper engine and microphone are pushed to the backend on boot, reproducing the lost-engine bug exactly.

SpotAI v1.3.2

SpotAI v1.3.2

🐛 Bug Fixes

• 🪟 No More Console Window: whisper-cli now spawns with CREATE_NO_WINDOW, so dictation no longer flashes an ugly cmd window while the WAV is transcribed.
• ⏱️ Dynamic Transcription Timeout: The wait for Whisper now scales with the recording length (min 20s, cap 5min) instead of a fixed 9s that used to give up before local CPU transcription finished on longer captures. The native engine keeps its fast 9s.

♻️ Refactors

• 🧩 Pure Helpers + Unit Tests: formatCaptureTime/formatDuration (src/lib/format.ts) and getVoiceTranscriptionTimeout (src/lib/voiceTimeout.ts) were extracted from the spotlight UI into pure, tested modules — 11 new Vitest cases cover the locale mapping, mm:ss formatting and the timeout bounds (9s native / 20s floor / 5min cap).

SpotAI v1.3.1

SpotAI v1.3.1

✨ New Features

• 🕐 Voice Session Indicators: The recording bar shows when the capture started (localized started at HH:MM:SS) plus a live elapsed counter, and the "Transcribing…" bar shows the total processed duration — so old captures are easy to tell apart.
• ⚠️ Microphone Permission Warning: When the Windows native recognizer fails with "speech privacy policy was not accepted", Settings shows an actionable banner (with a friendly toast instead of the cryptic OS error) that clears automatically once a transcription succeeds.
• 🧹 Automatic WAV Cleanup: Recorded WAVs are deleted after transcription on every path (success AND failure), the native engine no longer writes an unused WAV at all, and a startup sweep removes stale voice artifacts older than 24h — the temp folder can never fill the disk.
• 🔄 Instant Releases: The Release workflow now publishes the GitHub release automatically when the build finishes (no manual draft step).

🐛 Bug Fixes

• 🎯 Ghost-Click Fix: A stale voice-transcribed error (e.g. the OS recognizer failing while the user is STILL recording) no longer hides the recording bar — the recording state now only resets when the event belongs to the current session.

SpotAI v1.3.0

SpotAI v1.3.0

Changes since v1.2.3:

📚 Documentation

• add CHANGELOG entry for v1.2.3

📦 Other

• v1.3.0: voice input with SAPI + Whisper, E2E tests, README docs