System Overview

Logician is built on four core packages that form a layered architecture.

Package layers

Package details

agent-core

The foundation layer. Handles:

• LLM backend (OpenAI-compatible HTTP client)
• Agent loop execution
• Configuration management
• Session persistence (SQLite)
• Compaction (context window management)
• Hook system
• Tool registry and execution

coding-agent

The orchestration layer. Handles:

• System prompt construction
• Skills loading and activation
• Session management (bookmarks, branching)
• Trust model (permission modes)
• MCP server integration
• Prompt guidelines and context files

agent-capabilities

The intelligence layer. Handles:

• Reasoners (ToT, SSR, Reflexion, etc.)
• Subagent delegation
• Todo/task management
• EOH (Evolution of Hints) engine
• Tool selection and execution strategies

tui

The presentation layer. Handles:

• Terminal UI rendering
• Input handling
• Streaming output
• State management
• Layout and theming

Data flow

User Input → TUI → Coding Agent → Agent Core → LLM Backend
                                                                ↓
User Display ← TUI ← Coding Agent ← Agent Core ← LLM Response ←