WingmanAI provides real-time audio transcription and interactive chat with a ChatGPT-powered bot, leveraging transcripts as an extensive memory. It's designed for users who want to build a persistent, context-aware conversational AI, acting as a powerful tool for managing and querying past conversations.

How It Works

The system captures both system audio and microphone input, transcribing it in real-time using the Whisper model. A key feature is its token-efficient memory management, where only the current transcript chunk is passed to ChatGPT. For richer context, saved transcripts are cross-referenced with a vector database, allowing the bot to access historical conversation data.

Quick Start & Requirements

• Install dependencies: pip install -r requirements.txt
• For CUDA support (recommended for real-time performance): pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117
• Prerequisites: ffmpeg installed, OpenAI API key.
• Run: python main.py
• Demo: https://github.com/e-johnstonn/wingmanAI/assets/30129211/6f9f8e09-f43e-47d5-87ae-ac5bc693963d

Highlighted Details

• Real-time transcription of system and microphone audio.
• ChatGPT integration for interactive querying of transcripts.
• Efficient memory management using current transcript chunks.
• Vector database for cross-referencing saved transcripts to provide richer context.
• Ability to save, load, and append conversations for long-term memory building.

Maintenance & Community

• Contributions are welcomed; users can open issues for queries or issues.
• The project is a modified version of SevaSk's "Ecoute" project.

Licensing & Compatibility

• License details are not explicitly stated in the README.
• Compatibility is limited to Windows.

Limitations & Caveats

CPU transcription is not real-time. The project is currently Windows-only. The default transcription model is "base," with "tiny" as an alternative for less capable hardware.