# System Requirements

## Hardware Requirements

- **CPU**: x86_64. Intel Core Ultra (Meteor Lake) or newer is recommended.
  Older Intel Core / Xeon processors will run the service but may be slower
  on OpenVINO inference paths.
- **Memory**: 16 GB RAM minimum. 32 GB recommended when running ASR and
  sentiment together, when using larger Whisper variants, or when keeping
  multiple sessions warm.
- **Disk**: 20 GB free SSD space recommended for model assets, the Hugging
  Face cache, temporary audio chunks, and per-session storage. NVMe is
  preferred for faster first-run model export.
- **GPU (optional)**: Intel integrated GPU (Meteor Lake or newer iGPU) or a
  supported discrete GPU exposed via `/dev/dri` for the OpenVINO `GPU`
  device path.
- **NPU (optional)**: Intel NPU on Meteor Lake or newer, exposed via a host
  NPU device node and used by setting `ACCEL_MOUNT_PATH` to that node
  (commonly `/dev/accel/accel0` on Meteor Lake systems), plus the
  `ZE_ENABLE_ALT_DRIVERS=libze_intel_npu.so` runtime variable.
- **Microphone (optional)**: ALSA-compatible capture device if you intend
  to list devices via `GET /devices` or pass `/dev/snd` into the container.

| Device                | Minimum              | Recommended                                                                                         |
| --------------------- | -------------------- | --------------------------------------------------------------------------------------------------- |
| CPU                   | x86_64               | Intel Core Ultra (Meteor Lake) or newer                                                             |
| Memory                | 16 GB RAM            | 32 GB RAM                                                                                           |
| Disk                  | 20 GB free SSD space | NVMe storage                                                                                        |
| GPU (optional)        | Not applicable       | Intel integrated GPU (Meteor Lake or newer iGPU) or a supported discrete GPU exposed via `/dev/dri` |
| NPU (optional)        | Not applicable       | Intel NPU on Meteor Lake or newer exposed via a host NPU device node (commonly `/dev/accel/accel0`)      |
| Microphone (optional) | Not applicable       | ALSA-compatible capture device with `/dev/snd` access when needed                                   |

## Software Requirements

### Operating System

- Ubuntu 22.04 LTS (validated) or a compatible Linux distribution with a
  recent kernel.
- For container deployment: Docker Engine and Docker Compose v2.
- For GPU acceleration on Linux: Intel/OpenVINO host GPU runtime
  (e.g. `intel-opencl-icd`, `level-zero`) installed on the host. This is a
  separate prerequisite from the Python dependencies.

### Host Packages (Standalone Run)

The standalone path additionally requires:

```bash
sudo apt-get update
sudo apt-get install -y ffmpeg alsa-utils libsndfile1
```

### Python

- Python 3.10 or newer.
- Dependencies installed from `requirements.txt`.

### Hugging Face Access (For Speaker Diarization)

If you plan to enable speaker diarization in `config.yaml`, you must have:

- A [Hugging Face account](https://huggingface.co/settings/tokens) with a personal access token (free).
- Acceptance of the [Pyannote speaker-diarization model license](https://huggingface.co/pyannote/speaker-diarization-community-1)
  on Hugging Face (one-time gate acceptance required).
- The access token set as `HF_TOKEN` in `.env` or the environment before starting the service.

If speaker diarization is not enabled in `config.yaml`, `HF_TOKEN` is not required.

## Network Requirements

- Outbound internet access on first run to download model assets from
  Hugging Face, unless models are pre-staged under `models/` and the cache.
- Inbound access to TCP port `8010` (default) for API clients.