Browser Audio Capture
Give any AI agent ears for the browser. One Chrome extension captures audio from any tab — meetings, YouTube, podcasts, webinars, courses, earnings calls — and streams it to your AI pipeline.
Why Use This
Your AI agent can't hear anything happening in your browser. This skill fixes that. Capture audio from any Chrome tab and stream it to your agent — no API keys, no OAuth, no per-platform integrations.
Use cases: meeting summaries, YouTube/podcast notes, competitive intel from earnings calls, auto-notes from online courses, customer call analysis — anything that plays audio in a browser tab.
Works with any AI agent — Claude, ChatGPT, OpenClaw, LangChain, CrewAI, or your own. If your agent can run shell commands or receive HTTP, it gets browser audio.
Prerequisites
Chrome with remote debugging:
CODEBLOCK0
Python 3.9+ with aiohttp: INLINECODE0
Quick Start
CLI (any agent that can exec)
CODEBLOCK1
Chrome Extension (one-click, persistent)
- 1.
chrome://extensions/ → Developer mode → Load unpacked → INLINECODE2 - Join a meeting → click Percept icon → Start Capturing
- Close popup — capture continues in background
Supported Platforms
Google Meet • Zoom (web) • Microsoft Teams • Webex • Whereby • Around • Cal.com • Riverside • StreamYard • Ping • Daily.co • Jitsi • Discord — plus any future platform that runs in a browser.
Audio Output
Streams to http://127.0.0.1:8900/audio/browser as JSON:
CODEBLOCK2
Configure endpoint in scripts/extension/offscreen.js (PERCEPT_URL). Point it at Whisper, Deepgram, NVIDIA Riva, or any transcription service.
Troubleshooting
- - No tabs: Chrome needs INLINECODE6
- Button won't click: Remove + re-add extension (MV3 caches aggressively)
- Audio not arriving: Check receiver on port 8900. Extension sends to INLINECODE7
浏览器音频捕获
为任何AI代理赋予浏览器的听觉能力。 一个Chrome扩展程序即可捕获任意标签页的音频——会议、YouTube、播客、网络研讨会、课程、财报电话会议——并将其流式传输到您的AI管道中。
为何使用此功能
您的AI代理无法听到浏览器中发生的任何内容。此技能解决了这一问题。从任意Chrome标签页捕获音频并流式传输到您的代理——无需API密钥、无需OAuth、无需针对各个平台的集成。
使用场景: 会议摘要、YouTube/播客笔记、财报电话会议的竞争情报、在线课程的自动笔记、客户通话分析——任何在浏览器标签页中播放音频的内容。
适用于任何AI代理——Claude、ChatGPT、OpenClaw、LangChain、CrewAI或您自己的代理。只要您的代理能运行shell命令或接收HTTP请求,就能获取浏览器音频。
前置条件
启用远程调试的Chrome:
bash
macOS
/Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome \
--remote-debugging-port=9222 --user-data-dir=$HOME/.chrome-debug-profile &
Python 3.9+ 并安装aiohttp:pip install aiohttp
快速开始
命令行(任何可执行命令的代理)
bash
列出标签页——会议用🎙️标记
python3 -m browser_capture.cli tabs
自动检测并捕获会议标签页
python3 -m browser_capture.cli capture
持续监控模式
python3 -m browser_capture.cli watch --interval 15
停止
python3 -m browser_capture.cli stop
Chrome扩展程序(一键安装,持久运行)
- 1. chrome://extensions/ → 开发者模式 → 加载已解压的扩展程序 → scripts/extension/
- 加入会议 → 点击Percept图标 → 开始捕获
- 关闭弹出窗口——捕获在后台持续进行
支持的平台
Google Meet • Zoom(网页版)• Microsoft Teams • Webex • Whereby • Around • Cal.com • Riverside • StreamYard • Ping • Daily.co • Jitsi • Discord——以及任何未来在浏览器中运行的平台。
音频输出
以JSON格式流式传输到http://127.0.0.1:8900/audio/browser:
json
{
sessionId: browser_1709234567890,
audio: ,
sampleRate: 16000,
format: pcm16,
tabUrl: https://meet.google.com/...,
tabTitle: 每周站会
}
在scripts/extension/offscreen.js中配置端点(PERCEPT_URL)。可指向Whisper、Deepgram、NVIDIA Riva或任何转录服务。
故障排除
- - 无标签页:Chrome需要--remote-debugging-port=9222参数
- 按钮无法点击:移除并重新添加扩展程序(MV3缓存机制较为激进)
- 音频未到达:检查端口8900上的接收器。扩展程序发送至/audio/browser