Crush vs Kimi Code CLI: A Comprehensive Comparison Series
This series provides a detailed, module-by-module comparison of two AI coding assistant CLI projects:
- Crush (
C:\GitHub\crush) — written in Go, built on the Charmbracelet framework - Kimi Code CLI (
C:\GitHub\kimi-cli) — written in Python, developed by Moonshot - Performance, stability, standalone deployment → Choose Crush
- IDE integration, flexible extensibility, Kimi services → Choose Kimi Code CLI
Table of Contents
1. Overview — project positioning, tech stacks, directory structure 2. Architecture — layered architecture, concurrency models, state management 3. Core Modules — Agent / Session / Message / Tool 4. Tool System — built-in tools, parameter definitions, MCP integration, security policies 5. Configuration Management — config hierarchy, formats, hot reloading 6. Prompt System — template engine, agent inheritance, Skills system 7. File Operations — read/write/edit/search, security policies 8. Command Handling — CLI framework, slash commands, ACP/IDE integration 9. LLM Integration — provider abstraction, streaming, context compression 10. Extension Mechanisms — Skills / MCP / agent inheritance / subagent systems 11. Performance Comparison — startup time, memory footprint, concurrency performance 12. Conclusions & Recommendations — selection guidance, improvement directions
Key Findings
| Dimension | Crush | Kimi Code CLI | |-----------|-------|---------------| | Language | Go | Python | | Startup speed | ~50ms | ~400ms | | Memory usage | ~80MB | ~250MB | | IDE integration | None | ACP protocol | | LSP support | Full | None | | Agent inheritance | Not supported | Supported |