Find your next open-source project

Search by name, purpose, or keyword — or paste a GitHub repository URL

joeseesun/subcast

Media & Design

Desktop app that turns overseas podcasts and YouTube videos into bilingual subtitles with AI summaries and Q&A.

147
19
TypeScript
AGPL-3.0
2 days old

What it is

SubCast is a desktop player for podcasts and YouTube channels aimed at Chinese-speaking learners of English-language content. It subscribes to YouTube channels, Apple Podcasts, Xiaoyuzhou or any RSS feed, fetches new episodes, and turns them into sentence-by-sentence bilingual subtitles. Alongside the transcript it produces an AI guide (one-line summary, timeline, key points, suggested segments) and a chat interface that answers questions from the full transcript, with clickable timestamps that jump back to the source audio.

The app is built with Tauri 2 (Rust) and React/TypeScript. Network requests run on the Rust side so that a proxy configured in settings is applied consistently and CORS is avoided; data is stored in SQLite, and on-device recognition runs in a child process so a crash does not take down the main app (architecture notes).

Getting started

  1. Download the installer for your platform from the official download page or the latest GitHub release. Builds are available for macOS (Apple silicon and Intel), Windows 10/11, and Linux (AppImage, deb). macOS packages are signed and notarized.
  2. On Windows the installer is not code-signed yet; the first launch triggers SmartScreen, so choose "More info → Run anyway" as documented in the README (download section).
  3. Launch the app and add a source: paste a YouTube channel handle, channel or playlist URL, an Apple Podcasts or Xiaoyuzhou link, or any RSS feed. Single videos and episodes can also be opened directly.
  4. Open an episode. If an official transcript exists it is used; otherwise the app recognizes audio on-device with Qwen3-ASR 0.6B by default (about 1 GB download), or you can configure a cloud recognizer such as Doubao, SiliconFlow, Groq or OpenAI. Translation defaults to Google Translate and can be pointed at your own model.
  5. For AI guide and Q&A, add an API key under the model settings. DeepSeek, SiliconFlow, Kimi, Claude, OpenAI, OpenRouter, Gemini, Ollama, LM Studio and any OpenAI-compatible endpoint are supported. Keys are stored in the system keychain.

Constraints

  • Node.js 22+, Rust stable and CMake are required to build from source; Windows also needs LLVM and Linux needs libwebkit2gtk-4.1-dev libdbus-1-dev clang cmake. Dev and packaging commands are npm install, npm run tauri dev and npm run tauri build (development section).
  • On-device recognition keeps audio local but requires the model download; cloud recognition, translation and AI features send the relevant content to whichever provider you configure.
  • Shared subtitles are on by default: recognized subtitle text is uploaded anonymously to the project's server with a random install ID. Audio is not uploaded, and the feature can be disabled under Settings → General.
  • Watching YouTube from mainland China requires a proxy, configured under Settings → General or inherited from the system. When Google Translate is unavailable the app switches to a domestic relay.
  • The project is licensed AGPL-3.0; distributing a modified version or running it as a network service requires releasing the full source under the same license (license).
  • SubCast is a player and study tool; episode copyright remains with the original creators.

HotRepo's take

SubCast entered HotRepo's new-project list for TypeScript on 2026-11-18, when it was recorded at 103 stars with 3 stars gained and 14 forks in the roughly one-hour growth window beginning 03:46 UTC. That is a single, short observation window rather than a trend, and the repository has no recorded GitHub all-language daily rank. What stands out technically is the delivery shape: a Tauri desktop app with on-device ASR, per-episode transcripts stored in SQLite, and a user-supplied model for AI features — so the interesting question for readers is not star velocity but whether the local-first processing and shared-subtitle upload model fit their own privacy expectations.

Written by AI from the project's README, updated . The project's own docs are the source of truth.

You might also like

  1. ersinkoc/AI-Video-FactoryTurn a topic into a finished multilingual YouTube video using Claude agents, paid media APIs and Remotion rendering.TypeScript164
  2. abus-aikorea/voice-proA local WebUI that turns videos into transcribed, translated, cloned-voice dubs without cloud APIs.Python13,347
  3. SimoHypers/limusicA desktop YouTube Music client built with Rust and Tauri that streams ad-free audio through libmpv, for listeners who want a native player.Rust1,551