Vibe Coding 时代：如何选择和管理你的AI员工（二）搭建本地Claude Code模型服务，每人月立省200美金

Claude Code → Claude Router → GPT-OSS:20B (本地)
                        → Claude API (云端，备用)

# 安装Homebrew (如果没有)/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"# 安装Ollamabrew install ollama# 安装Python和依赖brew install pythonpip install torch torchvision torchaudiopip install transformers accelerate# 安装Claude Code Routernpm install -g @saoudrizwan/claude-code-router

# 启动Ollama服务ollama serve# 下载GPT-OSS:20B模型 (注意：需要32GB以上内存)ollama pull gpt-oss:20b# 验证模型下载ollama list

{  "Providers": [    {      "name": "ollama",      "api_base_url": "http://127.0.0.1:11434/v1/chat/completions",      "api_key": "ollama-local",      "models": ["gpt-oss:20b"]    }  ],  "Router": {    "default": "ollama,gpt-oss:20b"  }}

#!/bin/zsh
set -e

# === 配置参数 ===
MODEL="gpt-oss:20b"
API_KEY="ollama-local"
OLLAMA_HOST="127.0.0.1:11434"
CONFIG_FILE="$HOME/.claude-code-router/config.json"

echo "🔍 检查 Homebrew 安装..."
if ! command -v brew >/dev/null 2>&1; then
  echo "❌ Homebrew 未安装，请先执行："
  echo '/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"'
  exit 1
fi

echo "🔍 检查 Ollama..."
if ! command -v ollama >/dev/null 2>&1; then
  echo "📥 安装 Ollama..."
  brew install ollama
fi

echo "🔍 检查 Claude Code Router..."
if ! command -v ccr >/dev/null 2>&1; then
  echo "❌ 未检测到 ccr，请先安装 Claude Code Router"
  echo "👉 参考文档: https://github.com/saoudrizwan/claude-code-router"
  exit 1
fi

# === 启动 Ollama 服务 (默认 32k context) ===
echo "🚀 启动 Ollama 服务 (context_length=32768)..."
pkill ollama || true
OLLAMA_CONTEXT_LENGTH=32768 OLLAMA_API_KEY=$API_KEY ollama serve > /tmp/ollama.log 2>&1 &
sleep 2

# === 检查模型是否已拉取 ===
if ! ollama list | grep -q "$MODEL"; then
  echo "📥 拉取模型: $MODEL ..."
  ollama pull $MODEL
fi

# === 写入 Claude Code Router 配置 ===
echo "⚙️ 配置 Claude Code Router..."
mkdir -p "$(dirname $CONFIG_FILE)"
cat > $CONFIG_FILE <<EOF
{
  "Providers": [
    {
      "name": "ollama",
      "api_base_url": "http://$OLLAMA_HOST/v1/chat/completions",
      "api_key": "$API_KEY",
      "models": ["$MODEL"]
    }
  ],
  "Router": {
    "default": "ollama,$MODEL"
  }
}
EOF

# === 测试 Ollama API ===
echo "🧪 测试 Ollama API..."
curl -s http://$OLLAMA_HOST/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $API_KEY" \
  -d "{
    \"model\": \"$MODEL\",
    \"messages\": [{\"role\":\"user\",\"content\":\"hello with 32k context\"}]
  }" | jq '.choices[0].message.content'

# === 启动 Claude Code Router ===
echo "✅ 启动 Claude Code Router..."
exec ccr code

# 下载脚本
curl -o setup-claude-router.sh https://raw.githubusercontent.com/your-repo/setup-claude-router.sh

# 赋予执行权限
chmod +x setup-claude-router.sh

# 运行脚本
./setup-claude-router.sh

# 设置Metal后端加速
export PYTORCH_ENABLE_MPS_FALLBACK=1

# 优化内存使用
export OLLAMA_NUM_PARALLEL=1
export OLLAMA_MAX_LOADED_MODELS=1

# 启用Metal优化
ollama serve --metal

def route_request(request_text):
    # 简单的编程任务使用本地模型
    if is_simple_coding_task(request_text):
        return "local_gpt"
    
    # 复杂的推理任务使用Claude API
    if is_complex_reasoning(request_text):
        return "claude_api"
    
    # 默认使用本地模型
    return "local_gpt"