This project provides an AI-powered tool for transcribing and analyzing meeting recordings, targeting users who need to extract insights from audio and video content. It leverages Google's Gemini models to offer features like speaker diarization, meeting summaries, action item extraction, and video analysis, aiming to streamline post-meeting workflows.

How It Works

Offmute employs a multi-stage pipeline that first analyzes content by extracting screenshots and chunking audio. It then generates initial descriptions of visual and audio elements. The core transcription and diarization process uses context-aware audio chunk processing to identify speakers and maintain conversational flow, with real-time progress updates. For report generation, it utilizes a "Spreadfill" technique, creating a report structure with headings and then filling each section independently using the full context, ensuring coherence and detail while updating the report incrementally.

Quick Start & Requirements

• Primary install / run command: npx offmute <Meeting_Location> [options]
• Prerequisites: Node.js 14+, ffmpeg installed, Google Gemini API key.
• Links: Features, Quick Start, Installation, Usage, Advanced, How It Works

Highlighted Details

• Offers multiple processing tiers (First, Business, Economy, Budget, Experimental) utilizing different Gemini model combinations for cost/performance trade-offs.
• The experimental tier uses Gemini 2.5 Pro Preview with support for 65k token outputs.
• Supports custom instructions to guide AI analysis and allows saving intermediate processing files.
• Generates structured reports with key points, action items, and participant profiles, including video analysis for demos.

Maintenance & Community

• Created by Hrishi Olickel.
• Community support encouraged via GitHub repository starring.

Licensing & Compatibility

• License: Apache-2.0.
• Compatible with commercial use.

Limitations & Caveats

The project is described as an "experiment" and mentions "Maybe I went a little overboard though," suggesting potential scope creep or experimental stability. The "Experimental Tier" explicitly uses a preview model, which may have inherent instability or undocumented changes.