youtube git:20261014.71d43e0A

Immutable. This exact content never changes and is served at /api/v1/blob/707afea6ea8bb6b2.

---
name: youtube
description: >
  YouTube pack — acquire YouTube media (audio MP3 / video MP4 via yt-dlp)
  and publish finished videos to YouTube via the shared banodoco-social
  integration.
---

# YouTube

The YouTube pack acquires source media and publishes finished videos. It does
not edit or render media.

## Actions

| Action | Use it for |
|---|---|
| [`youtube.youtube_audio`](../actions/youtube_audio/STAGE.md) | Acquire an MP3 or MP4 from a search query or a direct URL supplied through `query`. |
| [`youtube.upload`](../actions/upload/STAGE.md) | Publish a reachable finished-video URL through the shared banodoco-social publisher. |

## When to use

- Use `youtube.youtube_audio` for source audio or video needed by downstream
  editing, understanding, or Training workflows.
- Use `youtube.upload` only when a finished video is already reachable at an
  HTTP(S) URL.

Do not use this pack to edit, render, or analyze media. Use the media,
video-editing, rendering, or understanding pack as appropriate.

## SDK quick-start

The v3 public SDK uses `kind="action"`. Outputs are Runtime-managed; do not
pass a local `out=` path or a made-up `url` input.

```python
import astrid.sdk as sdk
from astrid.sdk import AstridClient

with AstridClient.open_from_launcher() as client:
    result = sdk.invoke(
        "youtube.youtube_audio",
        kind="action",
        project="demo",
        client=client,
        inputs={"query": "cool synthwave mix", "mode": "audio"},
        wait=True,
    )
    if not result.ok:
        raise RuntimeError(result.error)
    outputs = client.tasks.list_managed_outputs(result.kernel_task_id)
    if not outputs.ok:
        raise RuntimeError(outputs.error)
    print(outputs.data)
```

Pass a direct URL through `query`, without changing the input shape:

```python
result = sdk.invoke(
    "youtube.youtube_audio",
    kind="action",
    project="demo",
    client=client,
    inputs={"query": "https://www.youtube.com/watch?v=...", "mode": "video"},
    wait=True,
)
```

`mode="video"` produces the named managed MP4 media output required by
Training. Audio mode produces MP3. Both modes retain the named
`media_manifest` receipt with the target, mode, actual output path/extension,
hash, bytes, and acquisition provenance. Callers should consume the returned
managed output records instead of assuming a local filename.

Publishing returns the shared publisher JSON result and no file artifact:

```python
result = sdk.invoke(
    "youtube.upload",
    kind="action",
    project="demo",
    client=client,
    inputs={
        "video_url": "https://example.com/hype.mp4",
        "title": "My Hype Video",
        "description": "Generated by Astrid",
        "tags": "hype,astrid",
        "privacy_status": "private",
        "made_for_kids": False,
    },
    wait=True,
)
if not result.ok:
    raise RuntimeError(result.error)
publisher_result = result.raw_result["payload"]["action_result"]
```

## Boundaries and requirements

- Acquisition uses `yt-dlp`; audio extraction also requires `ffmpeg`.
- YouTube acquisition stays behind the host-managed `youtube-media` broker;
  the pack does not change host policy, credentials, or workspace
  dependencies.
- Publishing delegates to the existing banodoco-social/Zapier integration and
  requires a reachable video URL. It does not upload local files itself.
- The cache policy is `none`; failed or cancelled actions are not successful
  media or publish results.

Read the linked action guide before admission and follow YouTube's terms and
copyright requirements.