Introduction to Omi
Relevant source files
The following files were used as context for generating this wiki page:
Omi is a fully open-source AI ecosystem designed to act as a "2nd brain." It captures real-time conversations and screen data to generate summaries, action items, and provide an AI chat interface that remembers historical interactions. The project encompasses specialized hardware wearables, mobile applications for iOS and Android, a macOS desktop application, and a robust Python-based backend.
The system is built to support continuous capture, transcription, and intelligent analysis across multiple platforms, trusted by over 300,000 professionals. It leverages a combination of local hardware processing and cloud-based AI services to provide seamless memory management and productivity tools.
Sources: README.md:1-15
System Architecture
The Omi architecture is divided into three primary tiers: User Devices (Wearables, macOS, Mobile), the Omi Backend (Python/FastAPI), and Third-party AI Services (Deepgram, LLMs).
The diagram above illustrates the multi-platform flow where audio data from wearables or applications is pushed to the backend for Voice Activity Detection (VAD) and Diarization before being processed by Large Language Models (LLMs).
Sources: README.md:95-120, AGENTS.md:75-100
Key Components
| Component | Responsibility | Technology Stack |
|---|---|---|
| macOS App | Desktop capture and UI | Swift, SwiftUI, Rust |
| Mobile App | Wearable pairing and mobile UI | Flutter (iOS/Android) |
| Backend API | Orchestration and processing | Python, FastAPI, Firebase |
| Firmware | Audio capture and transmission | nRF, Zephyr, C |
| Wearable | Continuous 24h+ audio capture | nRF5340, nRF7002 |
Sources: README.md:120-135, omi/hardware/consumer/README.md:40-50
Hardware and Firmware
Omi hardware includes the Omi Wearable and Omi Glass Dev Kit. The consumer wearable is an open-source design utilizing an nRF5340 dual-core Bluetooth LE SoC and an nRF7002 Wi-Fi 6 chip. It is designed for 24-hour continuous audio capture using dual TDK T5838 PDM microphones.
Firmware Architecture
The firmware is built on Zephyr RTOS with the nRF Connect SDK. It employs a dual-core architecture:
- Application Core (Cortex-M33): Handles main logic, OPUS audio encoding, and UI (LEDs/Haptics).
- Network Core (Cortex-M33): Manages the Bluetooth stack and radio communication.
The system supports secure Over-The-Air (OTA) updates via MCUboot, ensuring cryptographically signed firmware images can be safely updated through the mobile app.
Sources: omi/hardware/consumer/README.md:40-55, omi/firmware/BUILD_AND_OTA_FLASH.md:25-45, omi/firmware/BUILD_AND_OTA_FLASH.md:215-230
Desktop and Mobile Integration
The Omi ecosystem provides deep integration with desktop and mobile environments to ensure a comprehensive memory capture.
macOS Application
The desktop app utilizes a Swift frontend and a Rust backend. It features a unique "local bridge" for automation and interaction without synthetic mouse events.
- Path:
desktop/macos/ - Bundle ID:
com.omi.computer-macos - Features: Screen capture, realtime hub, and local API for agent access.
Sources: README.md:120-125, desktop/macos/AGENTS.md:5-20, desktop/macos/AGENTS.md:90-110
Mobile Application
Developed with Flutter, the mobile app acts as the primary interface for wearable management and memory viewing.
- Auth: Firebase Auth (Google/Apple Sign-in) with PKCE.
- l10n: Supports 49 locales for global accessibility.
- Native Bridge: Uses Pigeon for bidirectional communication between Dart and native iOS/Android stubs (handling recording, battery, and permissions).
Sources: app/AGENTS.md:10-25, app/AGENTS.md:40-50, app/AGENTS.md:100-115
Developer Tools and SDKs
Omi provides specialized tools for both human developers and AI agents to interact with the platform.
Omi CLI
The omi-cli is a Python-based SDK designed for terminal use and agent automation. It allows scoped access to Omi's four primary data structures:
- Memories: Facts and learnings.
- Conversations: Processed audio/text exchanges.
- Action Items: Tasks and follow-ups.
- Goals: Tracked progress metrics.
Sources: sdks/python-cli/README.md:1-25
Local Desktop API
The omi local command group allows developers to interact directly with the running macOS app to perform tasks like searching screen history or taking screenshots programmatically.
# Example Local API call via CLI
omi --json local search-screen "pricing page" --days 7 --app Safari
Sources: sdks/python-cli/README.md:100-125
Conclusion
Omi provides a vertically integrated AI memory stack, spanning from specialized low-power hardware to sophisticated GPU-accelerated backend processing. By maintaining an open-source approach across firmware, hardware designs, and software SDKs, Omi enables a transparent and extensible "second brain" architecture for professional productivity.
Sources: README.md:10-20, AGENTS.md:5-15