Interactive Mode
Launch the full-screen interactive TUI with push-to-talk, live hardware monitoring, and model management.Features:
- Push-to-talk voice recording (press SPACE)
- Text input mode
- Real-time hardware metrics
- Model hot-swapping
- Action browser
- Benchmark runner
- RAG document ingestion
SPACE- Start/stop voice recordingM- Models panel (browse, download, switch LLM/STT/TTS)A- Actions panel (enable/disable macOS actions)B- Benchmarks panelR- RAG panelD- Cleanup panel (delete unused models)T- Toggle tool call traceESC- Stop processing / close panel / quit
Voice Commands
Continuous voice mode - always listening for commands.RCLI continuously monitors your microphone, detects speech, processes commands, and responds via TTS. Press
Ctrl+C to stop.Example use case: Leave running in the background for hands-free Mac control.Text Commands
Execute a one-shot text command without voice input.Examples:RAG-enhanced queries:
Actions Management
List all 43 available macOS actions.Display all actions grouped by category (Productivity, Communication, Media, System, Window, Web/Nav, Clipboard).
Show detailed information for a specific action.
Execute an action directly with JSON arguments.Examples:
Model Management
Interactive model management dashboard for all AI models.Browse, download, and switch between models without restarting RCLI.
Browse, download, and switch TTS voices.Interactive picker for 6 TTS voices (Piper, KittenTTS, Kokoro, Matcha).
Guided LLM upgrade wizard.Download and switch to a larger, more capable language model (Qwen3.5 2B/4B, LFM2 2.6B).
Upgrade to Parakeet TDT high-accuracy STT.Downloads NVIDIA Parakeet TDT 0.6B v3 (~640MB) for ~1.9% WER (vs. Whisper’s ~5% WER).
RAG (Document Intelligence)
Index documents for retrieval-augmented generation.Examples:Supports PDF, DOCX, and plain text files. Creates a hybrid index (vector + BM25) in
~/Library/RCLI/index.Query indexed documents.Example:
Show RAG index information.
Benchmarking
Run comprehensive benchmarks across all subsystems.Suite options:
all- Run all benchmarks (default)stt- Speech-to-text onlyllm- Language model onlytts- Text-to-speech onlye2e- End-to-end pipelinetools- Tool-calling accuracy and latencyrag- RAG retrieval performancememory- Memory usage profiling
System Commands
Download default models (~1GB one-time setup).Downloads:
- LFM2 1.2B Tool (LLM)
- Zipformer (streaming STT)
- Whisper base.en (offline STT)
- Piper Lessac (TTS)
- Silero VAD
- Snowflake Arctic Embed S (embeddings)
Remove unused models to free disk space.Interactive tool to identify and delete models not currently selected.
Display engine information and installed models.Shows active models, hardware profile, and system configuration.
Test microphone audio levels.Records for 5 seconds and displays live audio level visualization. Useful for debugging microphone issues.
Help
Display command reference.
Exit Codes
0- Success1- Error (missing models, initialization failure, invalid arguments)
Related
Options
Command-line flags and parameters
Environment
Environment variables and config files