Updating default behavior

Fix errors

Fix

Fixing errors

Author: kssgarcia

ARCHITECTURE.md

@@ -11,7 +11,7 @@ OneVox uses a model-centric architecture where the backend is automatically sele
 
 | Feature | whisper.cpp | ONNX Runtime |
 |---------|-------------|--------------|
-| **Build** | Default | `--features onnx` |
+| **Build** | Default | Default (included) |
 | **Selection** | Auto (GGML models) | Auto (ONNX/Parakeet models) |
 | **Stability** | Production-ready | Experimental |
 | **Speed** | 50-200ms | Varies by model |
@@ -104,7 +104,7 @@ let transcription = model.transcribe(&audio_samples, 16000)?;
 
 **Build:**
 ```bash
-cargo build --release --features onnx
+cargo build --release  # ONNX support included by default
 ```
 
 **Implementation:** `src/models/onnx_runtime.rs` (571 lines)
@@ -195,11 +195,11 @@ pub trait ModelRuntime: Send + Sync {
 [model]
 # Backend auto-detected from model_path
 # - GGML models (ggml-*) use whisper.cpp
-# - Parakeet/ONNX models use ONNX Runtime (requires --features onnx)
+# - Parakeet/ONNX models use ONNX Runtime (included by default)
 
 model_path = "ggml-base.en"      # English-only (whisper.cpp)
 # model_path = "ggml-base"       # Multilingual, 99+ languages (whisper.cpp)
-# model_path = "parakeet-ctc-0.6b"  # ONNX model (requires --features onnx)
+# model_path = "parakeet-ctc-0.6b"  # ONNX model (included by default)
 
 # Device selection
 device = "auto"  # auto, cpu, gpu
@@ -225,7 +225,7 @@ preload = true
 - `ggml-large-v3` (2.9GB)
 - `ggml-large-v3-turbo` (1.6GB)
 
-*ONNX (requires --features onnx):*
+*ONNX (included by default):*
 - `parakeet-ctc-0.6b` - Multilingual, INT8 quantized
 
 **Switching models:**
@@ -237,11 +237,11 @@ preload = true
 
 ```toml
 [features]
-default = ["whisper-cpp", "overlay-indicator"]
+default = ["whisper-cpp", "onnx", "overlay-indicator"]
 
-# Model backends (mutually exclusive in practice, but can coexist)
-whisper-cpp = ["whisper-rs"]                # Native whisper.cpp (recommended)
-onnx = ["ort", "ort-sys", "ndarray"]        # ONNX Runtime (multilingual)
+# Model backends
+whisper-cpp = ["whisper-rs"]                # Native whisper.cpp (default)
+onnx = ["ort", "ort-sys", "ndarray"]        # ONNX Runtime (default)
 candle = ["candle-core", "candle-nn", "candle-transformers"]  # Pure Rust (experimental)
 
 # GPU acceleration (whisper-cpp only)
@@ -257,20 +257,17 @@ overlay-indicator = ["eframe", "winit"]  # Visual recording indicator
 
 **Build examples:**
 ```bash
-# Default (whisper.cpp + overlay)
+# Default (includes both whisper.cpp and ONNX)
 cargo build --release
 
-# With ONNX support
-cargo build --release --features onnx
-
-# Both backends available (larger binary)
-cargo build --release --features "whisper-cpp,onnx"
+# Whisper.cpp only (minimal build)
+cargo build --release --no-default-features --features whisper-cpp
 
 # GPU-accelerated whisper.cpp (macOS)
 cargo build --release --features metal
 
-# ONNX + TUI
-cargo build --release --features "onnx,tui"
+# GPU-accelerated with ONNX
+cargo build --release --features "metal"
 ```
 
 ## Design Principles

CONTRIBUTING.md

@@ -47,8 +47,8 @@ cargo test
 6. **Link issues** - Reference any related issues
 
 **For model backend changes:**
-- Test both `cargo build --release` and `cargo build --release --features onnx`
-- Verify existing tests pass with both backends
+- Add appropriate tests for both whisper.cpp and ONNX models
+- Verify existing tests pass with both backends (both included by default)
 - Update ARCHITECTURE.md if behavior changes
 
 ## Areas for Contribution
@@ -118,14 +118,13 @@ cargo test --features onnx
 When contributing changes that affect model backends:
 
 ```bash
-# Test default backend (whisper.cpp)
+# Test default build (includes both whisper.cpp and ONNX)
 cargo build --release
 cargo test
 ./target/release/onevox daemon --foreground
 
-# Test ONNX backend (if applicable)
-cargo build --release --features onnx
-cargo test --features onnx
+# Test ONNX models specifically
+cargo test --release
 # Edit config: model_path = "parakeet-ctc-0.6b"
 ./target/release/onevox daemon --foreground
 

Cargo.toml

@@ -89,7 +89,7 @@ enigo = "0.6"        # Keyboard/mouse simulation for text injection
 libc = "0.2"
 
 [features]
-default = ["whisper-cpp", "overlay-indicator"] # Native whisper.cpp is the primary backend
+default = ["whisper-cpp", "onnx", "overlay-indicator"] # Include both backends by default
 
 # Model backends
 whisper-cpp = ["whisper-rs"]

DEVELOPMENT.md

@@ -10,15 +10,33 @@ cargo build --release
 
 ## Build
 
-### Default Backend (whisper.cpp)
+```bash
+# Debug build (includes ONNX by default)
+cargo build
+
+# Release build (includes ONNX by default)
+cargo build --release
 
-**First Build (macOS only):**
+# Minimal build (whisper.cpp only, no ONNX)
+cargo build --release --no-default-features --features whisper-cpp
+```
 
-macOS requires environment variables on first build:
+### ONNX Runtime Support
+
+ONNX support is **included by default** in all builds. This enables the Parakeet model and other ONNX models.
+
+**What's included:**
+- Downloads ONNX Runtime binaries automatically (~150MB) via `ort-sys`
+- Builds ONNX inference backend
+- Enables ONNX model support (Parakeet, etc.)
+- Increases binary size by ~30MB
 
+**Testing ONNX:**
 ```bash
-CC=clang CXX=clang++ SDKROOT=$(xcrun --show-sdk-path) MACOSX_DEPLOYMENT_TARGET=13.0 \
-  cargo build --release
+# ONNX is available by default
+./target/release/onevox config init
+# Edit config.toml: model_path = "parakeet-ctc-0.6b"
+./target/release/onevox daemon --foreground
 ```
 
 **Why?** whisper.cpp compiles from source and needs proper SDK paths.
@@ -139,7 +157,7 @@ src/
 ├── vad/                 # Voice Activity Detection
 ├── models/              # Transcription models
 │   ├── whisper_cpp.rs   # whisper.cpp backend (default)
-│   ├── onnx_runtime.rs  # ONNX Runtime backend (--features onnx)
+│   ├── onnx_runtime.rs  # ONNX Runtime backend (default)
 │   ├── whisper_candle.rs # Pure Rust backend (experimental)
 │   └── runtime.rs       # ModelRuntime trait
 ├── platform/            # Platform-specific
@@ -160,7 +178,7 @@ scripts/                 # Installation and packaging scripts
 
 **Core:**
 - `whisper-rs` - Native whisper.cpp bindings (default backend)
-- `ort` + `ort-sys` - ONNX Runtime bindings (optional, `--features onnx`)
+- `ort` + `ort-sys` - ONNX Runtime bindings (default, included in all builds)
 - `handy-keys` - Global hotkey detection
 - `cpal` - Cross-platform audio
 - `enigo` - Text injection

INSTALLATION.md

@@ -49,7 +49,7 @@ OneVox is available in two build configurations:
 - **Latency**: Varies by model
 - **Models**: Parakeet, custom ONNX models
 
-**Installation**: Build from source with `--features onnx` flag (see [Build from Source](#build-from-source) below)
+**Installation**: Build from source (ONNX support included by default, see [Build from Source](#build-from-source) below)
 
 ---
 
@@ -408,7 +408,7 @@ After building, edit your config file to select a model:
 # Backend is auto-detected from model_path
 model_path = "ggml-base.en"         # English-only (whisper.cpp)
 # model_path = "ggml-base"          # Multilingual (whisper.cpp, 99+ languages)
-# model_path = "parakeet-ctc-0.6b"  # ONNX model (requires --features onnx build)
+# model_path = "parakeet-ctc-0.6b"  # ONNX model (included by default)
 
 device = "auto"  # auto, cpu, gpu
 preload = true

QUICKREF.md

@@ -100,7 +100,7 @@ sample_rate = 16000
 # Model identifier (backend auto-detected from path)
 model_path = "ggml-base.en"         # English-only, ~142MB
 # model_path = "ggml-base"          # Multilingual (99+ languages)
-# model_path = "parakeet-ctc-0.6b"  # ONNX (requires --features onnx)
+# model_path = "parakeet-ctc-0.6b"  # ONNX (included by default)
 
 device = "auto"  # auto, cpu, gpu
 preload = true   # Load model at startup
@@ -109,7 +109,7 @@ preload = true   # Load model at startup
 **Available Models:**
 - **English-only**: `ggml-tiny.en` (75MB), `ggml-base.en` (142MB), `ggml-small.en` (466MB), `ggml-medium.en` (1.5GB)
 - **Multilingual**: `ggml-tiny` (75MB), `ggml-base` (142MB), `ggml-small` (466MB), `ggml-medium` (1.5GB), `ggml-large-v2/v3` (2.9GB), `ggml-large-v3-turbo` (1.6GB)
-- **ONNX**: `parakeet-ctc-0.6b` (multilingual, INT8, requires `--features onnx` build)
+- **ONNX**: `parakeet-ctc-0.6b` (multilingual, INT8, included by default)
 
 Multilingual models automatically detect the spoken language. Backend is auto-selected based on model name.
 

README.md

@@ -76,11 +76,11 @@ cargo build --release
 - Alternative models (Parakeet CTC, etc.)
 - INT8 quantization for faster inference
 - ~250MB memory usage
-- Requires `--features onnx` flag
+- Included by default (no special flags needed)
 
 ```bash
-# Build with ONNX support
-cargo build --release --features onnx
+# Build (includes ONNX support by default)
+cargo build --release
 ```
 
 Backend selection is automatic based on model choice (see Configuration below).
@@ -159,7 +159,7 @@ OneVox uses a model-centric architecture where the backend is automatically sele
 - Alternative models with INT8 quantization
 - CPU-optimized inference
 - ~250MB memory usage
-- Requires `--features onnx` build flag
+- Included by default in all builds
 
 **Model Selection:**
 ```toml
@@ -168,7 +168,7 @@ OneVox uses a model-centric architecture where the backend is automatically sele
 # Backend is auto-detected from model_path
 model_path = "ggml-base.en"      # Uses whisper.cpp, English-only
 # model_path = "ggml-base"       # Uses whisper.cpp, multilingual (auto-detect language)
-# model_path = "parakeet-ctc-0.6b"  # Uses ONNX Runtime (requires --features onnx)
+# model_path = "parakeet-ctc-0.6b"  # Uses ONNX Runtime (included by default)
 device = "auto"                   # or "cpu", "gpu"
 preload = true
 ```

config.example.toml

@@ -109,7 +109,7 @@ adaptive = true
 #     - ggml-large-v3 (~2.9GB, 99+ languages)
 #     - ggml-large-v3-turbo (~1.6GB, 99+ languages, faster)
 #
-# Available ONNX models (requires --features onnx build):
+# Available ONNX models (included by default):
 #   - parakeet-ctc-0.6b (multilingual, 100+ languages, 15-25x RT, INT8 quantized)
 #
 # Backend auto-detection:

src/daemon/lifecycle.rs

@@ -150,12 +150,32 @@ impl Lifecycle {
                             break;
                         }
                         Err(e) => {
+                            let error_msg = e.to_string();
+
+                            // Check if this is a model-related error (missing model file)
+                            let is_model_error = error_msg.contains("Model file not found")
+                                || error_msg.contains("Model not found")
+                                || error_msg.contains("Download GGML models")
+                                || error_msg.contains("Model download incomplete");
+
                             if retry_count == 0 {
                                 error!("Failed to create dictation engine: {}", e);
-                                error!("⚠️  This is usually a permission issue. Please grant:");
-                                error!("   1. Input Monitoring permission");
-                                error!("   2. Accessibility permission");
-                                error!("   Then restart: launchctl kickstart -k gui/$(id -u)/com.onevox.daemon");
+
+                                // Only show permission hints for non-model errors
+                                if !is_model_error {
+                                    error!("⚠️  This is usually a permission issue. Please grant:");
+                                    error!("   1. Input Monitoring permission");
+                                    error!("   2. Accessibility permission");
+                                    error!("   Then restart: launchctl kickstart -k gui/$(id -u)/com.onevox.daemon");
+                                }
+                            }
+
+                            // Don't retry for model errors - they won't fix themselves
+                            if is_model_error {
+                                error!("❌ Cannot start without a valid model");
+                                error!("   Daemon will continue running but dictation won't work");
+                                error!("   Download a model and restart the daemon");
+                                break;
                             }
 
                             retry_count += 1;