GenAgent

A taskbar AI assistant for macOS/iOS that can be invoked anytime to help with programming tasks and general assistance.

What it does

GenAgent runs in your taskbar and provides instant AI assistance through:

• Multiple AI agents that can work on different tasks simultaneously
• File operations, command execution, and git integration
• Web search and content fetching
• Persistent memory and conversation history
• Voice input/output capabilities

Tech Stack

Backend (Rust)

• tokio - Async runtime
• nom - Parser combinators
• serde + Protocol Buffers - Serialization
• uniffi - Rust-Swift FFI bridge
• Custom fork of genai crate for multi-provider AI support

Frontend (Swift)

• SwiftUI + Combine
• Native macOS/iOS integration
• Real-time streaming UI updates

AI Providers

• OpenAI, Anthropic (Claude), Google (Gemini)
• Groq, Cohere, Ollama (local models)

Audio

• Deepgram (transcription)
• OpenAI (text-to-speech)
• Hotword detection

Architecture Highlights

• Tool-based system: Extensible architecture where agents use tools (read, write, exec, git, search, memory, thought) to accomplish tasks
• Custom protocol: Delimiter-based message protocol using ◊tool markers for reliable parsing
• Sandboxed execution: Working directory isolation for security
• Cross-platform FFI: Universal binary support for Apple Silicon/Intel, iOS devices/simulators
• Streaming responses: Real-time response handling with backpressure management
• Persistent state: Crash-resistant agent and conversation storage

Project Structure

genagent/
├── src/                    # Rust core
│   ├── agent/             # Agent management
│   ├── tools/             # Tool implementations  
│   ├── audio/             # Audio processing
│   └── proto/             # Protocol buffers
├── AgentUI/               # Swift UI
├── Generated/             # FFI bindings
└── pkg/rust-genai/        # AI providers

Key Features

• Multi-agent task management with independent execution contexts
• Parser combinator-based command parsing for reliability
• Protocol buffer serialization for efficient storage
• Exponential backoff retry logic for API resilience
• Weak reference management to prevent UI memory cycles
• State machine-based agent lifecycle management

Use Cases

Originally designed as a general-purpose AI assistant that lives in your taskbar, GenAgent can:

• Write and modify code
• Execute shell commands
• Search documentation
• Manage git repositories
• Fetch web content
• Remember context across sessions
• Handle voice commands

The architecture makes it suitable for any task that benefits from AI assistance with system-level access.