agent-cli 0.74.0__tar.gz → 0.76.0__tar.gz

{agent_cli-0.74.0 → agent_cli-0.76.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agent-cli
-Version: 0.74.0
+Version: 0.76.0
 Summary: A suite of AI-powered command-line tools for text correction, audio transcription, and voice assistance.
 Project-URL: Homepage, https://github.com/basnijholt/agent-cli
 Author-email: Bas Nijholt <bas@nijho.lt>
@@ -83,7 +83,11 @@ Requires-Dist: pytest-mock; extra == 'test'
 Requires-Dist: pytest-timeout; extra == 'test'
 Requires-Dist: pytest>=7.0.0; extra == 'test'
 Provides-Extra: vad
-Requires-Dist: silero-vad-lite>=0.2.1; extra == 'vad'
+Requires-Dist: onnxruntime>=1.16.0; extra == 'vad'
+Provides-Extra: whisper-transformers
+Requires-Dist: fastapi[standard]; extra == 'whisper-transformers'
+Requires-Dist: torch>=2.0.0; extra == 'whisper-transformers'
+Requires-Dist: transformers>=4.30.0; extra == 'whisper-transformers'
 Provides-Extra: wyoming
 Requires-Dist: wyoming>=1.5.2; extra == 'wyoming'
 Description-Content-Type: text/markdown
@@ -424,7 +428,7 @@ Our installation scripts automatically handle all dependencies:
 |---------|---------|-----------------|
 | **[Ollama](https://ollama.ai/)** | Local LLM for text processing | ✅ Yes, with default model |
 | **[Wyoming Faster Whisper](https://github.com/rhasspy/wyoming-faster-whisper)** | Speech-to-text | ✅ Yes, via `uvx` |
-| **[`agent-cli server whisper`](docs/commands/server/whisper.md)** | Speech-to-text (alternative) | ✅ Built-in, `pip install "agent-cli[whisper]"` |
+| **[`agent-cli server whisper`](docs/commands/server/whisper.md)** | Speech-to-text (alternative) | ✅ Built-in, `pip install "agent-cli[faster-whisper]"` |
 | **[Wyoming Piper](https://github.com/rhasspy/wyoming-piper)** | Text-to-speech | ✅ Yes, via `uvx` |
 | **[Kokoro-FastAPI](https://github.com/remsky/Kokoro-FastAPI)** | Premium TTS (optional) | ⚙️ Can be added later |
 | **[Wyoming openWakeWord](https://github.com/rhasspy/wyoming-openwakeword)** | Wake word detection | ✅ Yes, for `assistant` |
@@ -512,12 +516,13 @@ agent-cli install-extras rag memory vad
 
   • rag - RAG proxy server (ChromaDB, embeddings)
   • memory - Long-term memory proxy (ChromaDB)
-  • vad - Voice Activity Detection (silero-vad)
+  • vad - Voice Activity Detection (Silero VAD via ONNX)
   • audio - Local audio recording/playback
   • piper - Local Piper TTS engine
   • kokoro - Kokoro neural TTS engine
   • faster-whisper - Whisper ASR for CUDA/CPU
   • mlx-whisper - Whisper ASR for Apple Silicon
+  • whisper-transformers - Whisper ASR via HuggingFace transformers (safetensors)
   • wyoming - Wyoming protocol for ASR/TTS servers
   • server - FastAPI server components
   • speed - Audio speed adjustment
@@ -534,7 +539,8 @@ agent-cli install-extras rag memory vad
 
 ╭─ Arguments ────────────────────────────────────────────────────────────────────────────╮
 │   extras      [EXTRAS]...  Extras to install: rag, memory, vad, audio, piper, kokoro,  │
-│                            faster-whisper, mlx-whisper, wyoming, server, speed, llm    │
+│                            faster-whisper, mlx-whisper, whisper-transformers, wyoming, │
+│                            server, speed, llm                                          │
 ╰────────────────────────────────────────────────────────────────────────────────────────╯
 ╭─ Options ──────────────────────────────────────────────────────────────────────────────╮
 │ --list  -l        Show available extras with descriptions (what each one enables)      │
@@ -630,8 +636,8 @@ the `[defaults]` section of your configuration file.
 ```toml
 [defaults]
 # llm_provider = "ollama"  # 'ollama', 'openai', or 'gemini'
-# asr_provider = "wyoming" # 'wyoming' or 'openai'
-# tts_provider = "wyoming" # 'wyoming', 'openai', or 'kokoro'
+# asr_provider = "wyoming" # 'wyoming', 'openai', or 'gemini'
+# tts_provider = "wyoming" # 'wyoming', 'openai', 'kokoro', or 'gemini'
 # openai_api_key = "sk-..."
 # gemini_api_key = "..."
 ```
@@ -643,7 +649,7 @@ the `[defaults]` section of your configuration file.
 **Workflow:** This is a simple, one-shot command.
 
 1.  It reads text from your system clipboard (or from a direct argument).
-2.  It sends the text to a local Ollama LLM with a prompt to perform only technical corrections.
+2.  It sends the text to your configured LLM provider (default: Ollama) with a prompt to perform only technical corrections.
 3.  The corrected text is copied back to your clipboard, replacing the original.
 
 **How to Use It:** This tool is ideal for integrating with a system-wide hotkey.

{agent_cli-0.74.0 → agent_cli-0.76.0}/README.md

@@ -334,7 +334,7 @@ Our installation scripts automatically handle all dependencies:
 |---------|---------|-----------------|
 | **[Ollama](https://ollama.ai/)** | Local LLM for text processing | ✅ Yes, with default model |
 | **[Wyoming Faster Whisper](https://github.com/rhasspy/wyoming-faster-whisper)** | Speech-to-text | ✅ Yes, via `uvx` |
-| **[`agent-cli server whisper`](docs/commands/server/whisper.md)** | Speech-to-text (alternative) | ✅ Built-in, `pip install "agent-cli[whisper]"` |
+| **[`agent-cli server whisper`](docs/commands/server/whisper.md)** | Speech-to-text (alternative) | ✅ Built-in, `pip install "agent-cli[faster-whisper]"` |
 | **[Wyoming Piper](https://github.com/rhasspy/wyoming-piper)** | Text-to-speech | ✅ Yes, via `uvx` |
 | **[Kokoro-FastAPI](https://github.com/remsky/Kokoro-FastAPI)** | Premium TTS (optional) | ⚙️ Can be added later |
 | **[Wyoming openWakeWord](https://github.com/rhasspy/wyoming-openwakeword)** | Wake word detection | ✅ Yes, for `assistant` |
@@ -422,12 +422,13 @@ agent-cli install-extras rag memory vad
 
   • rag - RAG proxy server (ChromaDB, embeddings)
   • memory - Long-term memory proxy (ChromaDB)
-  • vad - Voice Activity Detection (silero-vad)
+  • vad - Voice Activity Detection (Silero VAD via ONNX)
   • audio - Local audio recording/playback
   • piper - Local Piper TTS engine
   • kokoro - Kokoro neural TTS engine
   • faster-whisper - Whisper ASR for CUDA/CPU
   • mlx-whisper - Whisper ASR for Apple Silicon
+  • whisper-transformers - Whisper ASR via HuggingFace transformers (safetensors)
   • wyoming - Wyoming protocol for ASR/TTS servers
   • server - FastAPI server components
   • speed - Audio speed adjustment
@@ -444,7 +445,8 @@ agent-cli install-extras rag memory vad
 
 ╭─ Arguments ────────────────────────────────────────────────────────────────────────────╮
 │   extras      [EXTRAS]...  Extras to install: rag, memory, vad, audio, piper, kokoro,  │
-│                            faster-whisper, mlx-whisper, wyoming, server, speed, llm    │
+│                            faster-whisper, mlx-whisper, whisper-transformers, wyoming, │
+│                            server, speed, llm                                          │
 ╰────────────────────────────────────────────────────────────────────────────────────────╯
 ╭─ Options ──────────────────────────────────────────────────────────────────────────────╮
 │ --list  -l        Show available extras with descriptions (what each one enables)      │
@@ -540,8 +542,8 @@ the `[defaults]` section of your configuration file.
 ```toml
 [defaults]
 # llm_provider = "ollama"  # 'ollama', 'openai', or 'gemini'
-# asr_provider = "wyoming" # 'wyoming' or 'openai'
-# tts_provider = "wyoming" # 'wyoming', 'openai', or 'kokoro'
+# asr_provider = "wyoming" # 'wyoming', 'openai', or 'gemini'
+# tts_provider = "wyoming" # 'wyoming', 'openai', 'kokoro', or 'gemini'
 # openai_api_key = "sk-..."
 # gemini_api_key = "..."
 ```
@@ -553,7 +555,7 @@ the `[defaults]` section of your configuration file.
 **Workflow:** This is a simple, one-shot command.
 
 1.  It reads text from your system clipboard (or from a direct argument).
-2.  It sends the text to a local Ollama LLM with a prompt to perform only technical corrections.
+2.  It sends the text to your configured LLM provider (default: Ollama) with a prompt to perform only technical corrections.
 3.  The corrected text is copied back to your clipboard, replacing the original.
 
 **How to Use It:** This tool is ideal for integrating with a system-wide hotkey.

{agent_cli-0.74.0 → agent_cli-0.76.0}/agent_cli/_extras.json

@@ -6,12 +6,16 @@
     ]
   ],
   "faster-whisper": [
-    "TODO: add description",
-    []
+    "Whisper ASR via CTranslate2",
+    [
+      "faster_whisper"
+    ]
   ],
   "kokoro": [
-    "TODO: add description",
-    []
+    "Kokoro neural TTS (GPU)",
+    [
+      "kokoro"
+    ]
   ],
   "llm": [
     "LLM framework (pydantic-ai)",
@@ -28,12 +32,16 @@
     ]
   ],
   "mlx-whisper": [
-    "TODO: add description",
-    []
+    "Whisper ASR for Apple Silicon",
+    [
+      "mlx_whisper"
+    ]
   ],
   "piper": [
-    "TODO: add description",
-    []
+    "Piper TTS (CPU)",
+    [
+      "piper"
+    ]
   ],
   "rag": [
     "RAG proxy (ChromaDB, embeddings)",
@@ -55,9 +63,16 @@
     ]
   ],
   "vad": [
-    "Voice Activity Detection (silero-vad)",
+    "Voice Activity Detection (Silero VAD via ONNX)",
+    [
+      "onnxruntime"
+    ]
+  ],
+  "whisper-transformers": [
+    "Whisper ASR via HuggingFace transformers",
     [
-      "silero_vad"
+      "transformers",
+      "torch"
     ]
   ],
   "wyoming": [

{agent_cli-0.74.0 → agent_cli-0.76.0}/agent_cli/_requirements/vad.txt

@@ -14,14 +14,20 @@ click==8.3.1
     #   typer-slim
 colorama==0.4.6 ; sys_platform == 'win32'
     # via click
+coloredlogs==15.0.1
+    # via onnxruntime
 dotenv==0.9.9
     # via agent-cli
+flatbuffers==25.12.19
+    # via onnxruntime
 h11==0.16.0
     # via httpcore
 httpcore==1.0.9
     # via httpx
 httpx==0.28.1
     # via agent-cli
+humanfriendly==10.0
+    # via coloredlogs
 idna==3.11
     # via
     #   anyio
@@ -30,6 +36,16 @@ markdown-it-py==4.0.0
     # via rich
 mdurl==0.1.2
     # via markdown-it-py
+mpmath==1.3.0
+    # via sympy
+numpy==2.3.5
+    # via onnxruntime
+onnxruntime==1.20.1
+    # via agent-cli
+packaging==25.0
+    # via onnxruntime
+protobuf==6.33.4
+    # via onnxruntime
 psutil==7.2.1 ; sys_platform == 'win32'
     # via agent-cli
 pydantic==2.12.5
@@ -40,6 +56,8 @@ pygments==2.19.2
     # via rich
 pyperclip==1.11.0
     # via agent-cli
+pyreadline3==3.5.4 ; sys_platform == 'win32'
+    # via humanfriendly
 python-dotenv==1.2.1
     # via dotenv
 rich==14.2.0
@@ -53,8 +71,8 @@ shellingham==1.5.4
     # via
     #   typer
     #   typer-slim
-silero-vad-lite==0.2.1
-    # via agent-cli
+sympy==1.14.0
+    # via onnxruntime
 typer==0.21.1
     # via agent-cli
 typer-slim==0.21.1

agent_cli-0.76.0/agent_cli/_requirements/whisper-transformers.txt

@@ -0,0 +1,256 @@
+# This file was autogenerated by uv via the following command:
+#    uv export --extra whisper-transformers --no-dev --no-emit-project --no-hashes
+annotated-doc==0.0.4
+    # via fastapi
+annotated-types==0.7.0
+    # via pydantic
+anyio==4.12.1
+    # via
+    #   httpx
+    #   starlette
+    #   watchfiles
+certifi==2026.1.4
+    # via
+    #   httpcore
+    #   httpx
+    #   requests
+    #   sentry-sdk
+charset-normalizer==3.4.4
+    # via requests
+click==8.3.1
+    # via
+    #   rich-toolkit
+    #   typer
+    #   typer-slim
+    #   uvicorn
+colorama==0.4.6 ; sys_platform == 'win32'
+    # via
+    #   click
+    #   tqdm
+    #   uvicorn
+dnspython==2.8.0
+    # via email-validator
+dotenv==0.9.9
+    # via agent-cli
+email-validator==2.3.0
+    # via
+    #   fastapi
+    #   pydantic
+fastapi==0.128.0
+    # via agent-cli
+fastapi-cli==0.0.20
+    # via fastapi
+fastapi-cloud-cli==0.10.1
+    # via fastapi-cli
+fastar==0.8.0
+    # via fastapi-cloud-cli
+filelock==3.20.3
+    # via
+    #   huggingface-hub
+    #   torch
+    #   transformers
+fsspec==2026.1.0
+    # via
+    #   huggingface-hub
+    #   torch
+h11==0.16.0
+    # via
+    #   httpcore
+    #   uvicorn
+hf-xet==1.2.0 ; platform_machine == 'aarch64' or platform_machine == 'amd64' or platform_machine == 'arm64' or platform_machine == 'x86_64'
+    # via huggingface-hub
+httpcore==1.0.9
+    # via httpx
+httptools==0.7.1
+    # via uvicorn
+httpx==0.28.1
+    # via
+    #   agent-cli
+    #   fastapi
+    #   fastapi-cloud-cli
+huggingface-hub==0.36.0
+    # via
+    #   tokenizers
+    #   transformers
+idna==3.11
+    # via
+    #   anyio
+    #   email-validator
+    #   httpx
+    #   requests
+jinja2==3.1.6
+    # via
+    #   fastapi
+    #   torch
+markdown-it-py==4.0.0
+    # via rich
+markupsafe==3.0.3
+    # via jinja2
+mdurl==0.1.2
+    # via markdown-it-py
+mpmath==1.3.0
+    # via sympy
+networkx==3.6.1
+    # via torch
+numpy==2.3.5
+    # via transformers
+nvidia-cublas-cu12==12.8.4.1 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via
+    #   nvidia-cudnn-cu12
+    #   nvidia-cusolver-cu12
+    #   torch
+nvidia-cuda-cupti-cu12==12.8.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-cuda-nvrtc-cu12==12.8.93 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-cuda-runtime-cu12==12.8.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-cudnn-cu12==9.10.2.21 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-cufft-cu12==11.3.3.83 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-cufile-cu12==1.13.1.3 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-curand-cu12==10.3.9.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-cusolver-cu12==11.7.3.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-cusparse-cu12==12.5.8.93 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via
+    #   nvidia-cusolver-cu12
+    #   torch
+nvidia-cusparselt-cu12==0.7.1 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-nccl-cu12==2.27.5 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-nvjitlink-cu12==12.8.93 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via
+    #   nvidia-cufft-cu12
+    #   nvidia-cusolver-cu12
+    #   nvidia-cusparse-cu12
+    #   torch
+nvidia-nvshmem-cu12==3.3.20 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-nvtx-cu12==12.8.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+packaging==25.0
+    # via
+    #   huggingface-hub
+    #   transformers
+psutil==7.2.1 ; sys_platform == 'win32'
+    # via agent-cli
+pydantic==2.12.5
+    # via
+    #   agent-cli
+    #   fastapi
+    #   fastapi-cloud-cli
+    #   pydantic-extra-types
+    #   pydantic-settings
+pydantic-core==2.41.5
+    # via pydantic
+pydantic-extra-types==2.11.0
+    # via fastapi
+pydantic-settings==2.12.0
+    # via fastapi
+pygments==2.19.2
+    # via rich
+pyperclip==1.11.0
+    # via agent-cli
+python-dotenv==1.2.1
+    # via
+    #   dotenv
+    #   pydantic-settings
+    #   uvicorn
+python-multipart==0.0.21
+    # via fastapi
+pyyaml==6.0.3
+    # via
+    #   huggingface-hub
+    #   transformers
+    #   uvicorn
+regex==2026.1.15
+    # via transformers
+requests==2.32.5
+    # via
+    #   huggingface-hub
+    #   transformers
+rich==14.2.0
+    # via
+    #   agent-cli
+    #   rich-toolkit
+    #   typer
+    #   typer-slim
+rich-toolkit==0.17.1
+    # via
+    #   fastapi-cli
+    #   fastapi-cloud-cli
+rignore==0.7.6
+    # via fastapi-cloud-cli
+safetensors==0.7.0
+    # via transformers
+sentry-sdk==2.49.0
+    # via fastapi-cloud-cli
+setproctitle==1.3.7
+    # via agent-cli
+setuptools==80.9.0 ; python_full_version >= '3.12'
+    # via torch
+shellingham==1.5.4
+    # via
+    #   typer
+    #   typer-slim
+starlette==0.50.0
+    # via fastapi
+sympy==1.14.0
+    # via torch
+tokenizers==0.22.2
+    # via transformers
+torch==2.9.1
+    # via agent-cli
+tqdm==4.67.1
+    # via
+    #   huggingface-hub
+    #   transformers
+transformers==4.57.5
+    # via agent-cli
+triton==3.5.1 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+typer==0.21.1
+    # via
+    #   agent-cli
+    #   fastapi-cli
+    #   fastapi-cloud-cli
+typer-slim==0.21.1
+    # via agent-cli
+typing-extensions==4.15.0
+    # via
+    #   anyio
+    #   fastapi
+    #   huggingface-hub
+    #   pydantic
+    #   pydantic-core
+    #   pydantic-extra-types
+    #   rich-toolkit
+    #   starlette
+    #   torch
+    #   typer
+    #   typer-slim
+    #   typing-inspection
+typing-inspection==0.4.2
+    # via
+    #   pydantic
+    #   pydantic-settings
+urllib3==2.3.0
+    # via
+    #   requests
+    #   sentry-sdk
+uvicorn==0.40.0
+    # via
+    #   fastapi
+    #   fastapi-cli
+    #   fastapi-cloud-cli
+uvloop==0.22.1 ; platform_python_implementation != 'PyPy' and sys_platform != 'cygwin' and sys_platform != 'win32'
+    # via uvicorn
+watchfiles==1.1.1
+    # via uvicorn
+websockets==15.0.1
+    # via uvicorn

{agent_cli-0.74.0 → agent_cli-0.76.0}/agent_cli/core/vad.py

@@ -1,24 +1,130 @@
-"""Voice Activity Detection using Silero VAD for speech segmentation."""
+"""Voice Activity Detection using Silero VAD for speech segmentation.
+
+Uses ONNX model with pure numpy inference (no torch dependency).
+"""
 
 from __future__ import annotations
 
 import logging
+import urllib.request
 from collections import deque
+from pathlib import Path
+from typing import TYPE_CHECKING
 
 from agent_cli import constants
 
-try:
+if TYPE_CHECKING:
     import numpy as np
-    from silero_vad_lite import SileroVAD
-except ImportError as e:
-    msg = (
-        "silero-vad-lite is required for the transcribe-daemon command. "
-        "Install it with: `pip install agent-cli[vad]` or `uv sync --extra vad`."
-    )
-    raise ImportError(msg) from e
+    from numpy.typing import NDArray
 
 LOGGER = logging.getLogger(__name__)
 
+# Silero VAD model URL and cache location
+SILERO_VAD_VERSION = "v6.2"
+SILERO_VAD_URL = (
+    f"https://github.com/snakers4/silero-vad/raw/{SILERO_VAD_VERSION}"
+    "/src/silero_vad/data/silero_vad.onnx"
+)
+SILERO_VAD_CACHE = Path.home() / ".cache" / "agent-cli" / f"silero_vad_{SILERO_VAD_VERSION}.onnx"
+
+
+def _get_model_path() -> Path:
+    """Get path to Silero VAD ONNX model, downloading if needed."""
+    if SILERO_VAD_CACHE.exists():
+        return SILERO_VAD_CACHE
+
+    LOGGER.info("Downloading Silero VAD model...")
+    SILERO_VAD_CACHE.parent.mkdir(parents=True, exist_ok=True)
+
+    # Download with progress
+    urllib.request.urlretrieve(SILERO_VAD_URL, SILERO_VAD_CACHE)  # noqa: S310
+    LOGGER.info("Silero VAD model downloaded to %s", SILERO_VAD_CACHE)
+    return SILERO_VAD_CACHE
+
+
+class _SileroVADOnnx:
+    """Pure numpy wrapper for Silero VAD ONNX model."""
+
+    def __init__(self, *, force_cpu: bool = True) -> None:
+        """Initialize the ONNX model session."""
+        import onnxruntime  # noqa: PLC0415
+
+        model_path = _get_model_path()
+
+        opts = onnxruntime.SessionOptions()
+        opts.inter_op_num_threads = 1
+        opts.intra_op_num_threads = 1
+
+        providers = ["CPUExecutionProvider"] if force_cpu else None
+        self._session = onnxruntime.InferenceSession(
+            str(model_path),
+            providers=providers,
+            sess_options=opts,
+        )
+        self._sample_rates = [8000, 16000]
+        self.reset_states()
+
+    def reset_states(self, batch_size: int = 1) -> None:
+        """Reset the internal state for a new audio stream."""
+        import numpy as np  # noqa: PLC0415
+
+        self._state = np.zeros((2, batch_size, 128), dtype=np.float32)
+        self._context = np.zeros(0, dtype=np.float32)
+        self._last_sr = 0
+        self._last_batch_size = 0
+
+    def __call__(self, audio: NDArray[np.floating], sample_rate: int) -> float:
+        """Process an audio chunk and return speech probability."""
+        import numpy as np  # noqa: PLC0415
+
+        if sample_rate not in self._sample_rates:
+            msg = f"Supported sample rates: {self._sample_rates}"
+            raise ValueError(msg)
+
+        # Expected samples per chunk
+        num_samples = 512 if sample_rate == 16000 else 256  # noqa: PLR2004
+        context_size = 64 if sample_rate == 16000 else 32  # noqa: PLR2004
+
+        if len(audio) != num_samples:
+            msg = f"Expected {num_samples} samples for {sample_rate}Hz, got {len(audio)}"
+            raise ValueError(msg)
+
+        # Ensure 2D: [batch, samples]
+        if audio.ndim == 1:
+            audio = audio.reshape(1, -1)
+
+        batch_size = audio.shape[0]
+
+        # Reset state if sample rate or batch size changed
+        if self._last_sr and self._last_sr != sample_rate:
+            self.reset_states(batch_size)
+        if self._last_batch_size and self._last_batch_size != batch_size:
+            self.reset_states(batch_size)
+        if not self._last_batch_size:
+            self.reset_states(batch_size)
+
+        # Initialize context if empty
+        if len(self._context) == 0:
+            self._context = np.zeros((batch_size, context_size), dtype=np.float32)
+
+        # Concatenate context with audio
+        x = np.concatenate([self._context, audio], axis=1)
+
+        # Run ONNX inference
+        ort_inputs = {
+            "input": x,
+            "state": self._state,
+            "sr": np.array(sample_rate, dtype=np.int64),
+        }
+        out, self._state = self._session.run(None, ort_inputs)
+
+        # Update context with last samples
+        self._context = x[:, -context_size:]
+        self._last_sr = sample_rate
+        self._last_batch_size = batch_size
+
+        return float(out[0, 0])
+
 
 class VoiceActivityDetector:
     """Silero VAD-based voice activity detection for audio segmentation.
@@ -56,7 +162,7 @@ class VoiceActivityDetector:
         )
 
         # Model and state
-        self._model = SileroVAD(sample_rate=sample_rate)
+        self._model = _SileroVADOnnx()
         self._pre_speech_buffer: deque[bytes] = deque(maxlen=pre_speech_windows)
         self._pending = bytearray()
         self._audio_buffer = bytearray()
@@ -74,7 +180,7 @@ class VoiceActivityDetector:
 
     def reset(self) -> None:
         """Reset VAD state for a new recording session."""
-        self._model = SileroVAD(sample_rate=self.sample_rate)
+        self._model.reset_states()
         self._pre_speech_buffer.clear()
         self._pending.clear()
         self._audio_buffer.clear()
@@ -84,8 +190,10 @@ class VoiceActivityDetector:
 
     def _is_speech(self, window: bytes) -> bool:
         """Check if audio window contains speech."""
+        import numpy as np  # noqa: PLC0415
+
         audio = np.frombuffer(window, dtype=np.int16).astype(np.float32) / 32768.0
-        prob = self._model.process(audio)
+        prob = self._model(audio, self.sample_rate)
         LOGGER.debug("Speech prob: %.3f, threshold: %.2f", prob, self.threshold)
         return prob >= self.threshold
 
@@ -136,7 +244,7 @@ class VoiceActivityDetector:
                     self._silence_samples = 0
                     self._speech_samples = 0
                     self._audio_buffer.clear()
-                    self._model = SileroVAD(sample_rate=self.sample_rate)
+                    self._model.reset_states()
             else:
                 # Not speaking - maintain rolling pre-speech buffer (auto-limited by deque maxlen)
                 self._pre_speech_buffer.append(window)

{agent_cli-0.74.0 → agent_cli-0.76.0}/agent_cli/example-config.toml

@@ -13,7 +13,10 @@
 [defaults]
 
 # --- Provider Selection ---
-# Select the default provider for each service ("ollama"/"openai"/"gemini" for LLM, "wyoming"/"openai" for ASR).
+# Select the default provider for each service.
+# LLM: "ollama", "openai", or "gemini"
+# ASR: "wyoming", "openai", or "gemini"
+# TTS: "wyoming", "openai", "kokoro", or "gemini"
 llm-provider = "ollama"  # "local" still works as a deprecated alias
 tts-provider = "wyoming"