MNS - Meeting Notes CLI

Video/Audio meeting analyzer — transcribes, summarizes, and generates comprehensive meeting notes using Gemini AI.

Features

• 🎬 Video/Audio Processing: Extract audio from video (mp4, mov, avi) or process audio directly (mp3, wav, m4a)
• 🤖 AI Transcription: Transcribe audio using Gemini API with high accuracy
• 📝 Meeting Notes: Generate comprehensive meeting notes with executive summary & detailed analysis
• 🌍 Multi-language: Auto-detect source language & translate to target language
• 🌐 Smart Terminology: Technical terms, code, keywords remain in English for clarity
• ⚡ Smart Processing: Auto-split long videos into chunks, parallel processing
• 🧹 Auto Cleanup: Automatically delete source files after processing
• 💰 Cost Effective: ~$0.45 per 3-hour meeting

Prerequisites

• Node.js ≥ 18
• FFmpeg (for audio extraction)
• Gemini API key (Get one here)

Installing FFmpeg

# macOS
brew install ffmpeg

# Ubuntu/Debian
sudo apt-get install ffmpeg

# Windows — Download from https://ffmpeg.org/download.html

Installation

# Clone and build
cd /Users/tandm/Documents/jjuidev/npm/mns-cli
bun install
bun run build

# Run locally
./dist/cli/cli.js --help

# Or install globally
npm link
mns --help

Quick Start

# Set Gemini API key
export GEMINI_API_KEY="your-api-key-here"

# Basic usage (auto-detect language, output in Vietnamese)
mns analyze meeting.mp4

# Process audio file directly
mns analyze recording.mp3

# With context
mns analyze meeting.mp4 --context "POC demo 31/03/2026"

# Output notes in English
mns analyze meeting.mp4 --target english

# Keep source files
mns analyze meeting.mp4 --keep

# Custom output directory
mns analyze meeting.mp4 --output ./notes

# Verbose logging
mns analyze meeting.mp4 --verbose

CLI Options

mns analyze <video> [options]

Arguments:
  video              Path to video or audio file (required)

Options:
  -l, --lang         Source language (default: english)
  -t, --target       Target language for output (default: vietnamese)
  -c, --context      Context about the meeting
  -o, --output       Output directory (default: ./output)
  -k, --keep         Keep source files after processing (default: false)
  -p, --parallel     Enable parallel processing (default: true)
  -m, --model        AI model to use (overrides config)
  -v, --verbose      Enable verbose logging (default: false)

Language Handling

Source Language (--lang): Language spoken in the meeting (default: english)

Target Language (--target): Language for notes output (default: vietnamese)

Smart Terminology: Technical terms (API, GraphQL, CI/CD, REST) remain in English for clarity.

Example (Vietnamese output):

## Tóm tắt điều hành

- Thảo luận về API REST design cho microservices
- Quyết định sử dụng GraphQL cho query phức tạp
- Triển khai CI/CD pipeline với GitHub Actions

Output Format

Meeting notes (Markdown) include:

• Overall Summary: Meeting overview in target language (technical terms in English)
• Executive Summary: 5-7 bullet points of key topics
• Attendees: Identified speakers
• Decisions: Key decisions with timestamps
• Action Items: Tasks with responsible parties
• Full Transcripts: Source & target language versions in expandable sections

Example output file: output/meeting-2026-04-01-meeting.mp4.md

Architecture

src/
├── cli.ts                      # Entry point
├── commands/
│   ├── analyze/                # Main pipeline
│   └── config/                 # Config get/set
├── providers/
│   ├── types.ts                # TranscribeProvider interface
│   ├── gemini-provider.ts      # Gemini implementation
│   └── index.ts                # Barrel export
├── services/
│   ├── video-processor.ts      # FFmpeg extraction
│   ├── audio-chunker.ts        # Split into 15-min chunks
│   ├── transcriber.ts          # Gemini transcription
│   ├── summarizer.ts           # Meeting notes generation
│   └── cleanup.ts              # Temp file cleanup
├── utils/
│   ├── config.ts               # Config priority chain
│   ├── output.ts               # Markdown formatting
│   └── ...
└── types/
    └── index.ts                # Interfaces

Pipeline: video → extract audio → chunk → parallel transcribe → generate notes → save markdown → cleanup

Cost

Using Gemini 2.5 Flash:

• 3-hour meeting: ~$0.45
• Audio input (345,600 tokens): $0.35
• Output (transcript + summary): $0.10

Configuration

Environment Variables

export GEMINI_API_KEY="your-key"
# or
export GOOGLE_API_KEY="your-key"

Priority Chain

CLI flag (--model) 
  → ENV (MNS_MODEL) 
  → Project config (.mns.json) 
  → Global config (~/.config/mns/config.json) 
  → Default (gemini-3.1-pro-preview)

Project Config (.mns.json)

{
  "env": { "GEMINI_API_KEY": "your-key" },
  "model": { "gemini": "gemini-2.5-flash" }
}

Global Config (~/.config/mns/config.json)

{
  "geminiModel": "gemini-2.5-flash"
}

Examples

Example 1: Basic

mns analyze team-standup.mp4

Example 2: Multilingual

# Vietnamese video → English notes
mns analyze client-call.mp4 --lang vietnamese --target english

Example 3: Long Meeting

mns analyze 3-hour-workshop.mp4 --verbose --parallel

Troubleshooting

FFmpeg not found: brew install ffmpeg (macOS) or sudo apt-get install ffmpeg (Ubuntu)

No API key: Set GEMINI_API_KEY or GOOGLE_API_KEY environment variable

Quota exceeded: Check billing at https://console.cloud.google.com/billing or use lighter model (--model gemini-2.0-flash)

Development

# Install dependencies
bun install

# Build
bun run build

# Development mode (watch)
bun run dev

# Clean build
bun run clean

Roadmap

• [ ] Web UI (FastAPI + React)
• [ ] Database storage (SQLite)
• [ ] Batch processing
• [ ] Custom summary templates
• [ ] Speaker identification

License

MIT — Author: jjuidev

Links

• Gemini API Docs
• FFmpeg Documentation