Overview
What this is
The compact read before the technical details.
A CLI-based coding agent that automates development workflows through intelligent tool orchestration and context-aware project management. Built with Python and FastAPI, it leverages AI agents to understand project structure, execute complex coding tasks, and manage files and dependencies. Features include multi-step reasoning, tool calling, and persistent context across sessions.
Capabilities
What it actually does
The useful parts, pulled out of the paragraph wall.
Multi-agent coordinator architecture spawns specialized subagents with background worker management (spawn, monitor, await, collect results) for parallel task execution
Dual UI system: React/Ink TypeScript frontend communicating via Unix socket IPC with Python bridge, plus Rich/prompt_toolkit classic fallback via --classic flag
Context compaction system automatically compresses long conversations to stay within model context limits while preserving critical information across session restarts backed by SQLite
Extensible skills system loads specialized definitions from .skills/ directories with YAML frontmatter, plus per-project customization via terminus.md instruction files
Implementation
Technology with jobs attached
Names are less useful than responsibilities. This is what each piece is doing.
Python 3.11+
Core agent loop with tool-calling architecture, async orchestration, and Pydantic models for LLM configs and tool schemas
React / Ink
TypeScript terminal UI with components (Banner, InputBox) communicating over Unix socket IPC to Python backend
SQLite
Persistent session history and conversation context storage
Groq / OpenRouter / Gemini
Multi-provider LLM inference for cost and performance flexibility
Ripgrep
Codebase search for pattern matching, dependency tracking, and issue location
Langfuse
Optional observability layer for tracing and monitoring agent runs