Bilibili / YouTube / Xiaoe transcripts and complete linux.do discussions to Obsidian
A privacy-first Chrome extension that turns platform-provided video captions and complete linux.do topics into structured Markdown, then writes them to Obsidian without switching applications.
下载最新版 / Download latest release
This project is a focused capture layer for an Obsidian and AI-assisted learning workflow:
platform evidence -> source-specific extraction -> structured Markdown -> Obsidian -> later AI processing
It currently has two source types:
- Video transcript: Bilibili, YouTube and available Xiaoe caption tracks.
- Discussion thread: complete linux.do topics, including floors and reply relationships.
It does not attempt to replace the official Obsidian Web Clipper. The official clipper remains the better general-purpose tool for arbitrary web pages, templates and page-level extraction. This extension is complementary: it uses source-specific interfaces when generic page extraction would miss caption tracks, unloaded discussion floors or reply structure.
- Bilibili AI/manual subtitle extraction.
- YouTube caption and transcript extraction.
- Experimental read-only detection of existing Xiaoe subtitle tracks.
- Smart language policy that keeps an English original as evidence and adds Chinese only as an auxiliary track when available.
- Single-speaker mode by default; optional two-person and multi-person context for later AI attribution.
- Optional timestamps, disabled by default.
- No full-video download and no local speech-recognition model.
- Reads the topic JSON and fetches missing post IDs, so the page does not need to be scrolled to the bottom.
- Three capture scopes: main post, main post plus topic-author replies, or all floors.
- Preserves floor number, author, date, source link, topic-author marker and reply relationship.
- Preserves headings, links, remote images, quotes, fenced code, lists and tables.
- Keeps links to omitted reply targets when a narrower capture scope is selected.
- Copy Markdown, download
.md, preview, or save through Obsidian Local REST API. - True background writes with exact read-back verification; Obsidian does not take focus.
- Identical notes are not duplicated. Conflicting paths receive a numeric suffix instead of being overwritten.
- New notes use
processing_status: pendingand point to a source-specific processing rule. - Settings and Local REST API credentials stay in Chrome local extension storage.
- Download and extract the latest ZIP from Releases.
- Open
chrome://extensions/and enable Developer mode. - Click Load unpacked and select the folder containing
manifest.json. - Open 保存设置, enter your own target folders and Local REST API details, then test the connection.
Do not move or delete the unpacked folder while Chrome is using it.
The normal obsidian:// URI can activate Obsidian even with a silent parameter. This extension therefore uses the open-source Obsidian Local REST API plugin.
- Install and enable Local REST API in Obsidian.
- Enable its loopback HTTP server, normally
http://127.0.0.1:27123. - Copy its API key into this extension's collapsed settings panel.
- Click 测试后台连接 once.
Requests are restricted to 127.0.0.1 or localhost. The key is never added to notes, logs or release files.
| Source | Supported | Boundary |
|---|---|---|
| Bilibili | Existing manual/AI subtitle tracks | No audio transcription when a track is absent |
| YouTube | Existing captions and transcript data | Preserves original-language evidence |
| Xiaoe | Existing TextTrack/VTT/JSON captions | No media download, ASR or access-control bypass |
| linux.do | Complete accessible topic and selected floors | Uses the current signed-in page session; no permission bypass |
Video notes use type: video-transcript and video-transcript-v1. Discussion notes use type: discussion-thread and discussion-thread-v1. Both begin as raw evidence with processing_status: pending.
The processing rules separate three actions:
- clean into a readable derivative without overwriting raw evidence;
- discuss the source without writing by default;
- distill a durable interpretation that links back to the source note.
See PRIVACY.md. Public source and release packages contain no personal Vault ID, API key, account, email, local path, browsing history, real transcript, private topic content or third-party token.
npm run check
npm testThe test suite covers source detection, transcript policies, linux.do scopes and reply links, Markdown metadata, loopback-only REST authentication, write/read-back verification and conflict handling.
- Platform UI and internal interfaces may change.
- Hardcoded subtitles burned into video frames require OCR and are not supported.
- Xiaoe support depends on whether the course publisher provides a readable caption track.
- linux.do extraction requires the current browser session to have access to the topic.
- Remote images remain remote links in the first release; attachment downloading is intentionally deferred.
- True background saving requires Obsidian to be running with Local REST API enabled.
- The extension does not bypass login, payment, access control or DRM.
The linux.do adapter is based on the MIT-licensed userscript linux.do 帖子保存到 Obsidian by zsq. See THIRD_PARTY_NOTICES.md for the exact source and reuse boundary.
This project is MIT licensed. See LICENSE.