> ## Documentation Index
> Fetch the complete documentation index at: https://docs.echophrase.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Requirements

> System requirements, GPU support, and the models Echophrase ships with

# Requirements

Echophrase runs on a wide range of hardware. The default model is tuned to feel
fast on a mid-range gaming GPU; a smaller "Lite" variant runs on machines with
no dedicated GPU at all.

## System

| Resource       | Minimum                                       | Recommended               |
| -------------- | --------------------------------------------- | ------------------------- |
| **OS**         | Windows 10 (22H2)                             | Windows 11                |
| **OS (macOS)** | macOS 13 (Apple Silicon) — *Metal in testing* | macOS 14+                 |
| **OS (Linux)** | Ubuntu 22.04 (CPU only)                       | Ubuntu 24.04              |
| **RAM**        | 4 GB                                          | 8 GB+                     |
| **Disk**       | 2 GB free                                     | 4 GB free                 |
| **GPU**        | None — CPU works                              | DirectX 12 GPU on Windows |

## GPU acceleration

Echophrase auto-detects the fastest available device on first launch — there
is nothing to configure.

| Platform    | Acceleration         | What you need                                     |
| ----------- | -------------------- | ------------------------------------------------- |
| **Windows** | DirectML             | Any DirectX 12 GPU (NVIDIA, AMD, Intel, Qualcomm) |
| **macOS**   | Metal *(in testing)* | Apple Silicon, macOS 13.3+                        |
| **Linux**   | CPU                  | GPU acceleration on the roadmap                   |

<Note>
  We dropped CUDA support to escape "requirements hell" — matching the right
  NVIDIA driver, the right CUDA Toolkit version, and the right cuDNN version
  turned every install into a multi-step yak-shave. DirectML is built into
  Windows, runs on every modern GPU regardless of vendor, and gives us the
  same wins for transcription workloads with none of the install pain. Native
  CPU and Apple Metal cover the rest.
</Note>

## Supported models

Pulled directly from the desktop app's model registry — these are the exact
models that ship today.

| Model                                | Parameters | Min VRAM            | Recommended GPU                 | Realtime speed |
| ------------------------------------ | ---------- | ------------------- | ------------------------------- | -------------- |
| **Parakeet-TDT 0.6B** *(default)*    | 0.6B       | 6 GB                | RTX 3080 or better              | 12×            |
| **Parakeet Lite**                    | 120M       | None (GPU optional) | Any modern GPU, even integrated | 2–5×           |
| **T5-Mini Grammar** *(post-process)* | Mini       | —                   | Any GPU                         | —              |

### Recommended NVIDIA card

For the **default Parakeet-TDT** model we recommend an **RTX 3080 or better**.
**Parakeet Lite** runs comfortably on integrated graphics or any older NVIDIA
card — you do not need a dedicated GPU to use Echophrase.

<Tip>
  **More models on the way.** We're evaluating additional Whisper variants and
  multilingual voice models. Want to vote on what ships next?
  [Join the Discord](https://discord.com/invite/Er6NUcx).
</Tip>