Overview

Mix is a sophisticated AI-powered development platform built with Go, designed as a clean architecture monolith with microservices patterns. The system provides both CLI and HTTP interfaces for interacting with multiple LLM providers while maintaining session-based conversations and real-time streaming capabilities.

High-Level Architecture

Database Connections: Both the Mix Application and AI Agent connect directly to the database for different purposes: Mix Application handles session management and user operations, while AI Agent requires direct access for real-time conversation history retrieval and message storage during multi-turn conversations.

Core Components

1. Application Core

The central orchestrator managing all system components:

• Session Management: Multi-session conversation tracking with working directories
• Message Handling: Multi-part message processing with tool call correlation
• Service Composition: Clean dependency injection of all system services

2. LLM Agent System

Sophisticated AI conversation management:

• Multi-turn Conversations: Context-aware dialog flows across sessions
• Tool Orchestration: Sequential tool execution with streaming events
• Plan Mode: Restricted mode for planning workflows
• Provider Abstraction: Unified interface across multiple LLM providers

3. Provider Layer

Multi-provider LLM support with advanced features:

• Supported Providers: Anthropic, OpenAI (GPT), Google (Gemini), AWS Bedrock, Local models
• Advanced Features: Reasoning blocks, function calling, context caching, token tracking
• Session-specific Caching: Optimized provider instances per conversation

4. Tool System

Extensible plugin architecture:

• Built-in Tools: File operations, shell execution, web search, media handling
• MCP Integration: Model Context Protocol for external tool plugins
• Permission-gated: All tool operations require user consent
• Real-time Execution: Live progress updates via Server-Sent Events

5. Data Architecture

Robust data persistence and management:

• SQLite or Turso: Local SQLite with WAL or cloud Turso database for distributed deployments
• Type-safe Queries: SQLC-generated queries for compile-time safety
• Session Isolation: Per-session directory structure with asset management
• Migration System: Goose-based schema evolution

Key Architectural Patterns

Event-Based Communication

Components communicate through events rather than direct calls:

• PubSub System: Broadcasts events to multiple subscribers
• Real-time Streaming: Server-Sent Events deliver live updates to frontend
• Tool Execution Events: Track progress and status during tool execution

Plugin System

Tools can be added without modifying core code:

• Standardized Interface: All tools implement the same execution pattern
• MCP Integration: External plugins via Model Context Protocol
• Runtime Discovery: New tools are detected and registered automatically

Security Architecture

Permission System

• User Consent Required: All tool operations need explicit approval
• Session-scoped: Permissions tied to specific conversations
• Audit Trail: Complete logging of all permission requests and grants

Data Security

• Flexible Storage: Data stored in local SQLite or secure cloud Turso database
• Session Isolation: File system isolation per session
• API Key Management: Secure storage and environment-based configuration

This architecture successfully balances sophistication with maintainability, providing a robust foundation for AI-powered development assistance while maintaining excellent performance and security characteristics.