Owl is a personal, always-on wearable AI system designed for continuous life logging and proactive assistance. It targets individuals interested in memory augmentation, productivity enhancement, and exploring novel human-computer interactions through multimodal data capture and local AI inference. The project aims to provide a transparent and open platform for developing and deploying such systems.

How It Works

Owl operates through a distributed architecture comprising wearable capture devices, an AI server, and presentation clients. Wearable devices (e.g., ESP32, Sony Spresense, Apple Watch) capture audio and location data, with plans for image and video. This data is streamed or chunked to a central AI server. The server processes the data using flexible inference options, supporting local models via Ollama or commercial APIs like GPT-4 and Whisper. It employs VAD-based endpointing for conversation segmentation and utilizes background processing queues for transcription, summarization, and data storage.

Quick Start & Requirements

• Server Setup: Instructions provided for macOS, Linux, Windows, and Docker.
• Capture Devices: Supports custom ESP platforms, Sony Spresense, and Apple Watch. Reference hardware "Bee" is available for community testing.
• Clients: Native iOS and web interfaces are available; Android support is planned.
• Dependencies: Python, FastAPI, Ollama (for local models), or API keys for commercial services. Specific hardware may require custom firmware.
• Resources: Server hosting and potential GPU for local inference.
• Documentation: Setup Guide, Technical Guide

Highlighted Details

• Multimodal Capture: Supports audio and location, with image and video planned.
• Flexible Inference: Integrates with Ollama for local LLMs/VLMs and commercial APIs (GPT-4, Deepgram).
• Wearable Support: Broad device compatibility, including custom hardware like the "Bee" device with 50-hour battery life.
• Data Flow Transparency: Detailed "Tour de Source" explains data path from capture to processed output.

Maintenance & Community

• Key Contributors: Ethan Sutin, Bart Trzynadlowski.
• Community: Primarily via Discord.
• Roadmap: Not explicitly detailed, but feature development is ongoing.

Licensing & Compatibility

• License: Not explicitly stated in the README. Compatibility for commercial use or closed-source linking is undetermined.

Limitations & Caveats

• The project is experimental, with ongoing development for vision/video capture and Android support.
• Conversation detection relies on VAD, which is noted as a "naive and unreliable heuristic."
• Security recommendations highlight potential data exposure if not hosted securely (e.g., over HTTP).