xin-code

README

Xin Code

 __  __ ___ _   _    ____ ___  ____  _____
 \ \/ /|_ _| \ | |  / ___/ _ \|  _ \| ____|
  \  /  | ||  \| | | |  | | | | | | |  _|
  /  \  | || |\  | | |__| |_| | |_| | |___
 /_/\_\|___|_| \_|  \____\___/|____/|_____|

AI-powered terminal coding agent, built with Go.

An open-source terminal AI agent that supports multiple model providers, real-time cost tracking, context visualization, and diff preview — all in a single binary with zero dependencies.

---

Why Xin Code?

Feature	Claude Code	Xin Code
Multi-model support	Anthropic only	Anthropic + OpenAI + any compatible endpoint
CC OAuth reuse	N/A	Read Claude Code's Keychain token, zero cost for subscribers
Real-time cost panel	Token count (small text)	Status bar with live cost in CNY/USD
Context visualization	Text warning "almost full"	Progress bar with color gradient (green -> yellow -> red)
Diff preview	Text-only confirmation	Colored unified diff panel before writing
Startup time	~500ms (Node.js)	~50ms (Go binary)
Distribution	npm install + Node runtime	brew install / single binary / go install

Features

• Agent Loop — Full agentic coding with tool use, permission system, auto-compact, and interrupt recovery
• Multi-Provider — Native Anthropic + OpenAI providers, plus any OpenAI-compatible endpoint via BASE_URL (OpenRouter, etc.)
• CC OAuth Reuse — Automatically reads Claude Code's OAuth token from macOS Keychain
• 20+ Built-in Tools — Read, Write, Edit, Bash, Glob, Grep, WebFetch, WebSearch, Agent, MCP, Task, and more
• 36 Slash Commands — /commit, /pr, /review, /cost, /compact, /skills, and more
• MCP Integration — Connect to any MCP server (stdio/SSE/HTTP)
• Skills & Plugins — Extend with custom skills (SKILL.md) and plugins (plugin.json)
• Hooks — preToolUse / postToolUse hooks to intercept or react to tool calls
• Interactive TUI — Built with Charm (Bubbletea + Lipgloss + Glamour)
• Session Management — Auto-save, resume, export to Markdown
• XINCODE.md — Project-level instructions file (like .cursorrules)

Installation

Homebrew (macOS / Linux)

brew install xincode-ai/tap/xin-code

Go Install

go install github.com/xincode-ai/xin-code@latest

Download Binary

# macOS / Linux
curl -fsSL https://github.com/xincode-ai/xin-code/releases/latest/download/xin-code_$(uname -s)_$(uname -m).tar.gz | tar xz
sudo mv xin-code /usr/local/bin/

# Verify
xin-code --version

Quick Start

1. Configure API Key

# Option A: Anthropic API Key
export ANTHROPIC_API_KEY=sk-ant-...

# Option B: OpenAI API Key
export OPENAI_API_KEY=sk-...

# Option C: Universal key
export XINCODE_API_KEY=your-key

# Option D: If you have Claude Code installed and logged in,
# Xin Code will automatically reuse your OAuth token (macOS only)

2. Launch

# Start in current directory
xin-code

# Check version
xin-code --version

3. Start Coding

> Help me refactor this function to use error wrapping

> /commit     # Auto-generate commit message
> /review     # Code review current changes
> /cost       # Check session cost
> /compact    # Compress context when running low

Configuration

Settings File

Global config: ~/.xincode/settings.json Project config: .xincode/settings.json

{
  "model": "claude-sonnet-4-6-20250514",
  "provider": "anthropic",
  "max_tokens": 16384,
  "permissions": {
    "mode": "default",
    "rules": [
      { "tool": "Bash(rm -rf *)", "behavior": "deny" },
      { "tool": "Bash(sudo *)",   "behavior": "deny" }
    ]
  },
  "cost": {
    "currency": "CNY",
    "budget": 10.0,
    "budget_action": "warn"
  }
}

Environment Variables

XINCODE_API_KEY          # Universal API key
ANTHROPIC_API_KEY        # Anthropic-specific
OPENAI_API_KEY           # OpenAI-specific
XINCODE_MODEL            # Default model
XINCODE_BASE_URL         # Custom API endpoint (OpenRouter, etc.)
XINCODE_PERMISSION_MODE  # bypass / acceptEdits / default / plan / interactive

Multi-Model Support

Providers

Provider	Models	Config
Anthropic	Claude Sonnet, Opus, Haiku	ANTHROPIC_API_KEY
OpenAI	GPT-4o, o1, o3, o4-mini	OPENAI_API_KEY
OpenRouter	200+ models	OPENAI_API_KEY + XINCODE_BASE_URL=https://openrouter.ai/api/v1
Any compatible	Via BASE_URL	XINCODE_API_KEY + XINCODE_BASE_URL

Auto-Routing

Model names are automatically routed to the correct provider:

• claude-*, sonnet-*, opus-*, haiku-* -> Anthropic
• gpt-*, o1-*, o3-*, o4-* -> OpenAI
• Everything else -> OpenAI (compatible mode)

Built-in Tools

Tool	Type	Description
Read	Read-only	Read file contents (images, PDFs supported)
Write	Write	Create or overwrite files
Edit	Write	Edit files with diff preview
Bash	Write	Execute shell commands (streaming output)
Glob	Read-only	File pattern matching
Grep	Read-only	Content search (ripgrep syntax)
WebFetch	Read-only	Fetch web page content
WebSearch	Read-only	Web search
Agent	Write	Spawn sub-agents with isolated context
MCP	Varies	Bridge to MCP server tools
AskUser	Read-only	Ask the user a question
Task	Write	Task management (create/get/list/update/stop)

Slash Commands

Category	Commands
Session	/help /session /resume /compact /clear /export /quit
Model & Config	/model /provider /config /login /logout /permissions /cost /status
Dev Workflow	/commit /pr /review /diff /plan /test /init /branch /refactor
Extensions	/mcp /skills /plugins /hooks /agents /team
System	/env /version /context /tips /upgrade /memory

MCP Integration

Configure MCP servers in settings.json:

{
  "mcp_servers": [
    {
      "name": "filesystem",
      "command": "mcp-server-filesystem",
      "args": ["/path/to/dir"]
    }
  ]
}

MCP tools are automatically registered as mcp__<server>__<tool> and follow the same permission system as built-in tools.

XINCODE.md

Create a XINCODE.md file in your project root to provide project-specific instructions to the AI agent. This is similar to .cursorrules or CLAUDE.md.

# My Project

## Overview
A web application built with Next.js and TypeScript.

## Code Style
- Use functional components with hooks
- Comments in Chinese
- Use pnpm as package manager

## Build & Test
- `pnpm dev` - Start dev server
- `pnpm test` - Run tests

Use /init to auto-generate one based on your project structure.

Skills

Create custom skills in ~/.xincode/skills/<name>/SKILL.md or .xincode/skills/<name>/SKILL.md:

---
name: my-skill
description: Description for AI to decide when to use
whenToUse: When the user asks about X
---

Skill content here (Markdown, injected as context when needed)

Hooks

Configure hooks in settings.json:

{
  "hooks": {
    "preToolUse": [
      {
        "match": "Bash",
        "command": "echo 'Running: $TOOL_INPUT'"
      }
    ],
    "postToolUse": [
      {
        "match": "",
        "command": "echo 'Tool $TOOL_NAME completed'"
      }
    ]
  }
}

• preToolUse hooks that exit non-zero will block tool execution
• postToolUse hooks run after tool completion
• Hooks timeout after 10 seconds

Permission Modes

Mode	Behavior
bypass	All tools auto-approved
acceptEdits	File read/write auto-approved, Bash requires confirmation
default	Read-only auto-approved, write tools require confirmation (recommended)
plan	Read-only allowed, write operations blocked
interactive	All tools require confirmation

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

MIT