# AnyContent Vault Importer

Paste a **Douyin / TikTok / WeChat Official Account / YouTube** link, and this plugin sends it to a small local backend that transcribes, scrapes, or OCRs the content into a clean Markdown note inside your vault.

One plugin, four platforms, AI-readable text output. No third-party services are called by the plugin itself — only your local backend.

## Features

| Source | What you get |
| --- | --- |
| Douyin / TikTok video | ASR transcript + no-watermark MP4 URL (optional local download) |
| Douyin image post (图文) | Per-image OCR'd Markdown + image URLs |
| WeChat Official Account article | Full article body in Markdown (text + inline image references) |
| YouTube video | Timestamped transcript + title / channel metadata |

Every imported note ships with a YAML frontmatter block (`title`, `source`, `author`, `platform`, `post_type`, `created`, `tags`, …) so it slots straight into Dataview, Bases, or your own queries.

## Architecture

```
Obsidian ── (HTTP, localhost only) ──> Python backend ── (ASR / OCR / scrape) ──> Markdown ──> vault
```

This plugin is the Obsidian side only. The backend is a separate FastAPI service that you run locally — see [vancoder4-cyber/anycontent-obsidian-backend](https://github.com/vancoder4-cyber/anycontent-obsidian-backend).

The split keeps the plugin tiny (no Python / no `ffmpeg` bundled) and lets the heavy lifting (audio extraction, model inference, etc.) live in a dependency stack that's natural for that work.

## Install

### From the community plugin store *(recommended once approved)*

Settings → Community plugins → Browse → search **AnyContent Vault Importer** → Install → Enable.

### Manual / BRAT (until approved)

1. Install [BRAT](https://github.com/TfTHacker/obsidian42-brat).
2. BRAT → "Add Beta plugin" → paste `https://github.com/vancoder4-cyber/anycontent-obsidian-importer`.
3. Enable the plugin in Settings → Community plugins.

### Run the backend

```bash
git clone https://github.com/vancoder4-cyber/anycontent-obsidian-backend
cd anycontent-obsidian-backend
uv sync
export API_KEY=sk-your-siliconflow-key   # free signup at https://cloud.siliconflow.cn/
uv run python web/app.py                  # listens on :8080
```

The API key is required for video ASR and image OCR (SiliconFlow's SenseVoice / Qwen-VL). WeChat articles and YouTube transcripts work without it.

## Usage

There are three ways to trigger an import:

1. **Ribbon icon** (the download arrow on the left sidebar) → paste link → Import.
2. **Command palette** → "Import URL to vault" or "Import URL from clipboard".
3. **Inbox folder** — turn on *Inbox auto-consume* in settings. Anything you write into `_anycontent-inbox/<name>.txt` is treated as share text and imported automatically. Useful for pairing with shell scripts, automations, or AI assistants that write files into your vault.

Imported notes land in your configured *Raw folder* (`raw/` by default) with a filename pattern you control via `{date}-{author}-{platform}-{title}`.

## Settings

| Setting | What it does |
| --- | --- |
| Backend URL | Where the local backend is listening. Defaults to `http://127.0.0.1:8080`. |
| API key | Override the backend's `API_KEY` per-vault (or leave blank to use the backend's env var). Stored locally only. |
| Raw folder | Vault-relative path where imported notes are written. |
| Filename pattern | Template with `{date} {author} {platform} {title} {videoId}` tokens. |
| Save no-watermark video locally | For Douyin / TikTok, also download the MP4 into `raw/<media>/` and embed it. |
| Media subfolder name | The subfolder used by the previous option. |
| Open note after import | Auto-open the new note when import finishes. |
| Inbox folder | Vault-relative folder watched for new files. |
| Inbox auto-consume | Toggle whether files dropped into the inbox are imported automatically. |

## Privacy / Network

- The plugin only makes HTTP calls to **your local backend URL**. It never connects directly to the internet from Obsidian.
- The backend, in turn, fetches the source page (e.g. `mp.weixin.qq.com`), and for video ASR / image OCR it forwards the audio / image to whichever provider you've configured your API key with (SiliconFlow by default).
- No analytics, telemetry, or remote logging in this plugin.
- API keys you paste into the settings tab are stored in Obsidian's plugin data file inside the vault. Treat the vault accordingly.

## Compatibility

Desktop-only (`isDesktopOnly: true`). The backend requires Python 3.10+ and runs anywhere `uv` / `pip` runs.

Minimum Obsidian: 1.4.0.

## Known limitations

- **YouTube transcripts** depend on YouTube exposing captions. Videos with captions disabled / blocked will still create a note (title + channel + author), with a clear "no transcript" message in the body.
- **WeChat image hotlinking** — `mmbiz.qpic.cn` images often refuse to load outside of WeChat due to Referer restrictions. The plugin embeds the URL anyway so the text content remains complete; viewing the images may require opening the original article.
- **WeChat articles behind login** cannot be extracted (the page returns a login wall).
- **Douyin OCR** can take ~5s per image on the default Qwen3-VL model. A 10-slide post takes ~1 minute.

## Contributing

Issues and PRs welcome. Source: <https://github.com/vancoder4-cyber/anycontent-obsidian-importer>.

## License

MIT — see `LICENSE`.

The backend repo carries its own license (Apache 2.0, inheriting from upstream `yzfly/douyin-mcp-server`).