Private, fully local bilingual meeting engine for your Mac.
私有、全本地的 Mac 双语会议引擎
English · 简体中文 · Try It Live · Website & Docs
JargonSlayer captures both sides of an English meeting at once — system audio and microphone, echo-cancelled and auto-tagged 我/对方 (macOS 26+). Local Whisper, Apple's on-device speech recognition, and on-device translation form the default pipeline — fully local, nothing leaves your Mac; cloud engines are strictly opt-in. A live bilingual transcript scrolls as people talk, and when the meeting ends one click files the bilingual summary into Obsidian — or copies it Notion-ready. Jargon still gets a short Chinese card, now live assist rather than the headline.
The product UI is Simplified Chinese — it is built for non-native English speakers (Chinese-speaking professionals and researchers first).
- macOS desktop app (Apple Silicon, recommended) — download the DMG from Releases. The first-run wizard installs local Whisper into the app's own directory (nothing downloads without consent; deleting the directory is a clean uninstall). System/app audio (macOS 14.4+) transcribes meetings running in native apps — Zoom, Teams, WeChat — directly, with no virtual audio device.
- Web app — try the hosted preview (built-in demo AI key, rate-limited), or self-host:
npm install && npm run build && npm startwith Node 20+. Details in the docs. Works in mobile browsers too. - iOS (beta) — available via TestFlight on request (get in touch); on-device system speech recognition and on-device system translation (Apple Translation, no key required), iOS 26+.
- Chrome extension (JargonSlayer Lite) — live captions and jargon cards in a side panel. Manual install (the supported path; no Web Store listing for now): download the zip from Releases, unzip, open
chrome://extensions, enable Developer mode, click "Load unpacked", pick the unzipped folder. Step-by-step guide.
- Dual capture — system audio and microphone recorded together, echo-cancelled (AEC) so your mic doesn't also pick up what's playing from your speakers; no virtual audio device (desktop, macOS 26+).
- Fully local by default — local Whisper, Apple on-device recognition and on-device translation run on your machine; with them, zero bytes leave your Mac. Cloud engines (bring your own key) are opt-in and labeled in the UI.
- Bilingual live transcript — English audio, Chinese translation, side by side as people talk.
- 我/对方 speaker attribution — dual capture tags your mic and the meeting audio as 我 (you) and 对方 (them) automatically; pyannote diarization (optional one-click install) breaks out individual speakers live or from an imported recording.
- Export to your vault — one click files a bilingual summary and full transcript translation into Obsidian, or copies it Notion-ready and opens Notion to paste; also Anki, Markdown, JSON, and webhooks.
- Jargon cards (live assist) — slang, idioms, metaphors, acronyms → short Chinese explanations as they're spoken; dictionary mode is default (1,000+ entries, fully local, no key needed), with an optional AI mode (Beta, bring your own key) one toggle apart.
- Learning center — personal glossary, flip-card review with spaced repetition, a word cloud, and known-word suppression, all in one place.
- Imports — audio/video files and URLs, transcribed locally in the browser or by the sidecar.
- Bit 🐉 — the pixel dragon perched on the status line. Click it. Click it three times fast.
Transparent by default, fully local when you choose. Local Whisper and system/app audio never leave your machine — the status line shows 「音频在本地处理」; every other engine and tier states exactly where audio and text go. The full data-path table lives on the privacy page.
Setup guides (API keys, speaker diarization, troubleshooting), the per-platform engine matrix, and the FAQ all live on the website: mianaz.github.io/jargonslayer · Docs (EN) · 文档（中文）
Issues, pull requests, forks, and modified versions are all welcome — the license exists to make that easy, not to guard the code. This is a best-effort side project, so reviews may be slow; forking ahead instead of waiting is a perfectly good answer. If you build on it, keep the copyright and license notices intact (the AGPL requires this), and a visible link back to this repository in whatever you publish is appreciated. Contributions you submit are accepted under the project's AGPL-3.0 license.
AGPL-3.0 © 2026 Miana Zeng. Free to use anywhere, including at work; if you modify it and redistribute or host it as a service, you must share your source under the same terms. Releases up to and including v0.3.0 were published under MIT and remain so.
This is a personal side project, maintained on a best-effort basis — no support, uptime, or fitness for any purpose is promised (see the license's warranty disclaimer).