v0.45.0

🎉 Gemini TTS Support

This release adds Google Gemini as a TTS provider, completing full Gemini provider parity across all three AI services (ASR, LLM, TTS).

✨ New Features

• Gemini TTS provider - Use Gemini's native text-to-speech across all voice commands:

  agent-cli speak "Hello world" --tts-provider gemini
  agent-cli chat --tts-provider gemini
  agent-cli assistant --tts-provider gemini
  agent-cli voice-edit --tts-provider gemini

• New CLI options: --tts-gemini-model and --tts-gemini-voice
• Default model: gemini-2.5-flash-preview-tts
• Available voices: Kore (default), Puck, Charon, Fenrir, and more

🔧 Improvements

• Consolidated PCM-to-WAV conversion into a single reusable function
• Documentation fixes for markdown formatting and emoji rendering

📚 Full Changelog

Full Changelog: v0.44.0...v0.45.0

v0.44.0

What's New

✨ Features

• Gemini ASR Support: Added Google Gemini as an ASR provider using native audio understanding capabilities (#177)
• Documentation Site: New Zensical-powered documentation site with auto-generated option tables (#157, #159, #162)

📚 Documentation

• Simplified Windows installation with native Windows support (#170)
• Added iOS Shortcut Guide to documentation (#165)
• Various improvements: GitHub-flavored admonitions, syntax highlighting, updated logo (#166, #169, #173-176)

🔧 Improvements

• Fixed VAD tests on Windows to prevent torch hang (#156)
• Cleaned up scripts directory and inlined zellij help text (#171, #172)
• DRY refactoring in docs generation (#164)

Full Changelog: v0.43.0...v0.44.0

v0.43.0

What's New

macOS Service Improvements

• Ollama now runs as a brew service (#155): Instead of manually managing ollama serve, Ollama now runs as a proper background service via brew services. The Zellij dashboard shows service status and control commands when running as a brew service.

• Whisper runs as a launchd service on Apple Silicon (#154): On ARM Macs, wyoming-mlx-whisper now runs as a native launchd service instead of a manual process, improving reliability and startup behavior.

Bug Fixes

• Fixed Windows CI test hang (#152): Resolved an issue where sounddevice's Pa_Initialize could hang during tests on Windows CI.

v0.42.0

🎙️ New Feature: Continuous Transcription Daemon

This release adds transcribe-daemon, a background service that continuously captures audio and automatically transcribes speech segments using voice activity detection (VAD).

Features

• Voice Activity Detection: Uses Silero VAD (neural network-based) for real-time speech/silence detection
• Pre-speech Buffer: Captures 300ms of audio before speech is detected to avoid missing word beginnings
• Automatic Segmentation: Segments audio based on configurable silence threshold (default 1s)
• Optional LLM Processing: Cleanup and formatting of transcriptions using existing prompts
• Audio Storage: Saves segments as MP3 files organized by date (~/.config/agent-cli/audio/YYYY/MM/DD/)
• JSON Lines Logging: Logs transcriptions with timestamps, role, raw/processed text, audio file paths
• Systemd Integration: Easy service installation for always-on transcription

Installation

# Install with VAD dependency
pip install 'agent-cli[vad]'
# or
uv tool install 'agent-cli[vad]'

Usage

# Basic daemon
agent-cli transcribe-daemon

# With custom role and silence threshold
agent-cli transcribe-daemon --role meeting --silence-threshold 1.5

# With LLM cleanup
agent-cli transcribe-daemon --llm --role notes

Full Changelog: v0.41.0...v0.42.0

v0.41.0

What’s Changed

• feat(memory): use JSON mode for fact reconciliation (#146) @basnijholt
• feat(memory): add list_all method to MemoryClient (#147) @basnijholt
• feat(chroma): add batched upsert to prevent embedding service overload (#145) @basnijholt

v0.40.0

What’s Changed

• perf(rag): use mtime-first checking for faster startup (#144) @basnijholt
• Update uv.lock (#143) @basnijholt

v0.39.1

What’s Changed

• docs: run proxy in foreground for visible logs (#142) @basnijholt
• docs: disable Open WebUI auth for instant demo access (#141) @basnijholt
• docs: add 'Try It Now' quick start examples (#136) @basnijholt
• fix: set default log level to INFO for proxies (#140) @basnijholt
• feat: add version and -v/--version CLI flag (#138) @basnijholt

v0.39.0

What’s Changed

• refactor: move OnnxCrossEncoder to shared core module (#137) @basnijholt
• docs: add high-level overviews to architecture docs (#135) @basnijholt
• [pre-commit.ci] pre-commit autoupdate (#133) @pre-commit-ci[bot]
• feat(rag): use XML tags for context formatting (#131) @basnijholt
• feat(rag): add confidence threshold and fallback prompt (#130) @basnijholt

v0.38.0

What's Changed

• feat(whisper): use MLX-based Whisper on macOS for Apple Silicon by @basnijholt in #123
• feat: add stop file mechanism for graceful shutdown on Windows by @basnijholt in #127
• refactor: simplify Windows process management with psutil by @basnijholt in #129

Full Changelog: v0.37.1...v0.38.0

v0.37.1

What’s Changed

• fix(docs): correct AHK v2 script and use SIGINT (#125) @basnijholt
• fix(docs): correct AutoHotkey v2 syntax in Windows installation guide (#124) @basnijholt