# Hildegard documentation

> Install Hildegard, pick a model for your GPU, direct shots and films, pair your computers, host a studio at a link, link Hugging Face and connect your AI apps over MCP.

Project page: https://dbaney.com/projects/hildegard/

## Install

On macOS and Linux:

```sh
curl -fsSL https://dbaney.com/hildegard/install.sh | sh
```

On Windows, in PowerShell (no admin rights):

```powershell
irm https://dbaney.com/hildegard/install.ps1 | iex
```

The installer checks every download against a SHA-256 written into it, puts everything in one folder in your home, sets up ComfyUI and PyTorch for your GPU, and starts Hildegard. The desktop app for Windows, macOS and Linux does the same in a window; download it from the [Hildegard site](https://dbaney.com/hildegard/#install).

Then:

```sh
hildegard setup     # picks a video model that fits your GPU and links your AI apps
hildegard open      # opens the Studio
```

## What your GPU runs

| GPU memory | Model |
| --- | --- |
| 24 GB or more (NVIDIA, or AMD on Linux) | LTX-2.5 distilled: video with sound, up to 1088p |
| 12 to 23 GB | Wan 2.2 5B: silent video at 720p |
| A Mac with Apple silicon | Renders on a paired computer with an NVIDIA or AMD GPU |
| Less, or none | Renders on another computer, or a GPU rented on Runpod |

LTX asks you to accept its license on Hugging Face first, so link your account with `hildegard hf login`. The token goes to your system keychain.

## Render on another computer

Put both computers on your Tailscale network and install Hildegard on both. On the one with the GPU, `hildegard choir invite` prints a join code; on the other, `hildegard choir join` and paste it. `hildegard choir list` shows the devices, and shots render there from then on.

## Host a studio at a link

```sh
hildegard host owner        # your name and a passphrase
hildegard host              # prints your link, inside your tailnet
hildegard host invite       # a one-time link for someone else
```

`hildegard host --funnel` puts the link on the public internet, which claude.ai and ChatGPT need to connect.

## Connect your AI apps

```sh
hildegard connect --all     # every AI app on this computer
```

In Claude Code, Hildegard is also a plugin: `/plugin marketplace add baney75/hildegard`, then `/plugin install hildegard@hildegard`. Linked apps get tools to direct shots, check takes, make films, speak in consented voices and run the make, check and improve loop.

## Fix problems

`hildegard doctor` checks every part and prints the command that fixes each problem. Every command takes `--json` for agents. The full reference is in the [README](https://github.com/baney75/hildegard#readme), and agents setting Hildegard up for someone should follow [INSTALL-FOR-AGENTS.md](https://github.com/baney75/hildegard/blob/main/INSTALL-FOR-AGENTS.md).