本篇文章主要介绍OpenClaw自定义任意云端模型的配置安装脚本及使用方法，主要作用为设置本地模型或自定义模型信息。 作者：任聪聪 日期：2026年3月9日
代码实例
#!/usr/bin/env python3
"""
Cloud Model Configuration Assistant for OpenClaw
自动配置任意云端大模型 (OpenAI, Anthropic, Gemini, DeepSeek, etc.) 到 OpenClaw。
"""
import json
import os
import sys
import urllib.request
import urllib.error
from datetime import datetime
from typing import Any, Dict, List, Optional, Tuple
# --- 配置常量 ---
DEFAULT_CONTEXT_WINDOW = 32768
DEFAULT_MAX_TOKENS = 8192
OPENCLAW_MIN_CONTEXT = 16000
# 预定义提供商模板
PROVIDER_TEMPLATES = {
"openai": {
"name": "OpenAI",
"base_url": "https://api.openai.com/v1",
"api_key_env": "OPENAI_API_KEY",
"api_type": "openai-completions",
"models_hint": "gpt-4o, gpt-4-turbo, gpt-3.5-turbo"
},
"anthropic": {
"name": "Anthropic (Claude)",
"base_url": "https://api.anthropic.com/v1", # OpenClaw 通常通过 adapter 处理，这里假设使用 openai-compatible 代理或原生支持
# 注意：如果 OpenClaw 原生支持 Anthropic，api_type 可能是 'anthropic'。
# 此处为了通用性，假设用户可能使用 openai-compatible 网关，或者脚本需特殊处理。
# 为简化，本脚本主要针对 OpenAI 兼容接口。若需原生 Anthropic，需确认 OpenClaw 具体协议。
# 这里我们假设 OpenClaw 支持通过 'openai-completions' 类型配合特定 header 或专门类型。
# *修正*：根据常见架构，若 OpenClaw 支持多协议，这里动态设置。
"api_type": "openai-completions", # 许多聚合层将 Anthropic 转为 OpenAI 格式，或使用特定 adapter
"models_hint": "claude-3-5-sonnet-20241022, claude-3-opus-20240229"
},
"google": {
"name": "Google Gemini",
"base_url": "https://generativelanguage.googleapis.com/v1beta/openai", # Gemini 开放了 OpenAI 兼容接口
"api_key_env": "GEMINI_API_KEY",
"api_type": "openai-completions",
"models_hint": "gemini-1.5-pro, gemini-1.5-flash"
},
"deepseek": {
"name": "DeepSeek",
"base_url": "https://api.deepseek.com/v1",
"api_key_env": "DEEPSEEK_API_KEY",
"api_type": "openai-completions",
"models_hint": "deepseek-chat, deepseek-coder"
},
"moonshot": {
"name": "Moonshot (Kimi)",
"base_url": "https://api.moonshot.cn/v1",
"api_key_env": "MOONSHOT_API_KEY",
"api_type": "openai-completions",
"models_hint": "moonshot-v1-8k, moonshot-v1-32k, moonshot-v1-128k"
},
"siliconflow": {
"name": "SiliconFlow (硅基流动)",
"base_url": "https://api.siliconflow.cn/v1",
"api_key_env": "SILICONFLOW_API_KEY",
"api_type": "openai-completions",
"models_hint": "Qwen/Qwen2.5-72B-Instruct, meta-llama/Llama-3.1-405B-Instruct"
},
"custom": {
"name": "Custom OpenAI-Compatible (vLLM, OneAPI, etc.)",
"base_url": "", # 用户输入
"api_key_env": "",
"api_type": "openai-completions",
"models_hint": "Any model served by your endpoint"
}
}
def print_banner():
print("=" * 70)
print("☁️ OpenClaw 云端模型配置助手")
print("=" * 70)
print()
def get_config_path() -> str:
possible_paths = [
os.path.expanduser("~/.openclaw/openclaw.json"),
os.path.expanduser("~/.moltbot/moltbot.json"),
]
env_path = os.environ.get("OPENCLAW_CONFIG_PATH")
if env_path:
return env_path
for path in possible_paths:
if os.path.exists(path):
return path
return possible_paths[0]
def load_config(path: str) -> Dict[str, Any]:
if not os.path.exists(path):
print(f"⚠️ 配置文件不存在: {path}，将创建新配置...")
return {"models": {"providers": {}}, "agents": {"defaults": {}}}
try:
with open(path, "r", encoding="utf-8") as f:
return json.load(f)
except Exception as e:
print(f"❌ 读取失败: {e}")
sys.exit(1)
def backup_config(path: str) -> str:
if not os.path.exists(path):
return ""
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
backup_path = f"{path}.backup.{timestamp}"
import shutil
shutil.copy2(path, backup_path)
print(f"✅ 已创建备份: {backup_path}")
return backup_path
def save_config(config: Dict, path: str) -> bool:
try:
os.makedirs(os.path.dirname(path), exist_ok=True)
with open(path, "w", encoding="utf-8") as f:
json.dump(config, f, indent=2, ensure_ascii=False)
return True
except Exception as e:
print(f"❌ 保存失败: {e}")
return False
def fetch_models_openai_compat(base_url: str, api_key: str) -> Tuple[List[Dict], Optional[str]]:
"""从兼容 OpenAI 格式的端点获取模型列表"""
models_url = f"{base_url.rstrip('/')}/models"
request = urllib.request.Request(models_url, method="GET")
request.add_header("Authorization", f"Bearer {api_key}")
try:
with urllib.request.urlopen(request, timeout=15) as response:
data = json.loads(response.read().decode("utf-8"))
models = data.get("data", []) if isinstance(data, dict) else data
# 过滤掉明显的 embedding 模型，除非用户特别需要，这里主要关注 LLM
llm_models = []
for m in models:
if isinstance(m, dict) and "id" in m:
mid = m["id"]
if any(k in mid.lower() for k in ["embed", "bge", "m3e"]):
continue
llm_models.append({"id": mid})
return llm_models, None
except urllib.error.HTTPError as e:
return [], f"HTTP 错误 {e.code}: {e.reason} (检查 API Key 或 URL)"
except urllib.error.URLError as e:
return [], f"网络错误: {e.reason}"
except Exception as e:
return [], str(e)
def get_api_key(provider_key: str, template: Dict) -> str:
env_var = template.get("api_key_env")
if env_var and env_var in os.environ:
key = os.environ[env_var]
if key:
print(f"🔑 从环境变量 {env_var} 自动获取 API Key")
return key
prompt_msg = f"请输入 {template['name']} 的 API Key: "
if env_var:
prompt_msg += f"(或设置环境变量 {env_var})\n> "
else:
prompt_msg += "\n> "
key = input(prompt_msg).strip()
if not key:
print("❌ API Key 不能为空。")
return get_api_key(provider_key, template)
return key
def select_provider() -> Tuple[str, Dict]:
print("📋 选择云服务提供商:")
keys = list(PROVIDER_TEMPLATES.keys())
for i, k in enumerate(keys):
name = PROVIDER_TEMPLATES[k]["name"]
print(f" [{i}] {name}")
while True:
choice = input("\n您的选择 (输入数字或名称): ").strip().lower()
if choice.isdigit():
idx = int(choice)
if 0 <= idx < len(keys):
return keys[idx], PROVIDER_TEMPLATES[keys[idx]]
elif choice in PROVIDER_TEMPLATES:
return choice, PROVIDER_TEMPLATES[choice]
print("⚠️ 无效输入，请重试。")
def build_model_config(model_id: str, context_window: int, cost_input: float, cost_output: float) -> Dict:
input_types = ["text"]
if "vl" in model_id.lower() or "vision" in model_id.lower() or "gemini" in model_id.lower():
input_types.append("image")
return {
"id": model_id,
"name": f"{model_id} (Cloud)",
"reasoning": "o1" in model_id.lower() or "deepseek-r" in model_id.lower(),
"input": input_types,
"cost": {
"input": cost_input,
"output": cost_output,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": context_window,
"maxTokens": min(DEFAULT_MAX_TOKENS, context_window // 4),
"compat": {"supportsDeveloperRole": True}
}
def main():
print_banner()
# 1. 选择提供商
provider_key, template = select_provider()
provider_id = f"cloud-{provider_key}"
print(f"\n⚙️ 配置提供商: {template['name']}")
# 2. 获取 Base URL
base_url = template["base_url"]
if provider_key == "custom":
base_url = input(f"请输入自定义 API Base URL (例如 http://localhost:8000/v1): ").strip()
if not base_url:
print("❌ URL 不能为空")
sys.exit(1)
else:
print(f" 默认 API 端点: {base_url}")
override = input(" 是否覆盖默认端点？(y/N): ").strip().lower()
if override == 'y':
base_url = input(" 输入新的 Base URL: ").strip()
# 3. 获取 API Key
api_key = get_api_key(provider_key, template)
# 4. 获取模型列表 (如果是 OpenAI 兼容格式)
available_models = []
error = None
# 尝试自动探测模型
if template["api_type"] == "openai-completions":
print(f"\n🔄 正在从 {base_url} 拉取模型列表...")
available_models, error = fetch_models_openai_compat(base_url, api_key)
if error:
print(f"⚠️ 自动探测失败: {error}")
print(" 将进入手动输入模式。")
elif not available_models:
print("⚠️ 未检测到模型列表 (可能接口不支持 /models 或返回空)。")
print(" 将进入手动输入模式。")
# 5. 确定要配置的模型
final_models = [] # List of Dicts: {'id': ..., 'ctx': ..., 'cost_in': ..., 'cost_out': ...}
if available_models:
print(f"✅ 发现 {len(available_models)} 个模型。")
print("\n📋 请选择要启用的模型 (可多选，用逗号分隔索引，或直接回车全选):")
for i, m in enumerate(available_models):
print(f" [{i}] {m['id']}")
selection = input("\n选择 (例如 0,2,5 或 回车全选): ").strip()
indices = []
if not selection:
indices = range(len(available_models))
else:
try:
indices = [int(x.strip()) for x in selection.split(',')]
except ValueError:
print("⚠️ 输入格式错误，默认全选。")
indices = range(len(available_models))
for i in indices:
if 0 <= i < len(available_models):
final_models.append(available_models[i]["id"])
else:
# 手动输入
print("\n✍️ 手动输入模型 ID (例如 gpt-4o, claude-3-5-sonnet-20241022):")
models_input = input("模型 ID (多个用逗号分隔): ").strip()
if not models_input:
print("❌ 至少需要配置一个模型。")
sys.exit(1)
final_models = [m.strip() for m in models_input.split(',')]
# 6. 配置全局参数 (上下文 & 成本)
print(f"\n🔧 全局参数配置 (可对单个模型微调，此处设默认值):")
ctx_input = input(f"默认上下文窗口 (回车默认 {DEFAULT_CONTEXT_WINDOW}): ").strip()
context_window = int(ctx_input) if ctx_input else DEFAULT_CONTEXT_WINDOW
if context_window < OPENCLAW_MIN_CONTEXT:
confirm = input(f"⚠️ 警告: {context_window} < {OPENCLAW_MIN_CONTEXT}。继续？(y/N): ").strip().lower()
if confirm != 'y':
sys.exit(0)
cost_in = input("默认输入成本 ($/1M tokens, 回车默认 0): ").strip()
cost_out = input("默认输出成本 ($/1M tokens, 回车默认 0): ").strip()
try:
c_in = float(cost_in) if cost_in else 0.0
c_out = float(cost_out) if cost_out else 0.0
except ValueError:
print("⚠️ 成本输入无效，设为 0。")
c_in, c_out = 0.0, 0.0
# 构建模型配置对象
model_configs = []
print("\n🏗️ 正在构建模型配置...")
# 如果有多个模型，可以让用户为每个模型单独确认，这里简化为批量应用默认值
# 高级用法可以在这里循环询问每个模型的特定 ctx/cost
for mid in final_models:
# 简单启发式调整某些知名模型的默认值
m_ctx = context_window
m_c_in, m_c_out = c_in, c_out
if "128k" in mid.lower() or "gemini-1.5" in mid.lower():
m_ctx = 131072
elif "32k" in mid.lower():
m_ctx = 32768
config_item = build_model_config(mid, m_ctx, m_c_in, m_c_out)
model_configs.append(config_item)
print(f" - {mid} (Context: {m_ctx})")
# 7. 选择主模型
if not model_configs:
print("❌ 没有模型被配置。")
sys.exit(1)
print(f"\n📋 选择默认主模型:")
for i, mc in enumerate(model_configs):
marker = " [推荐]" if i == 0 else ""
print(f" [{i}] {mc['id']}{marker}")
choice = input("您的选择 (回车默认第一个): ").strip()
try:
idx = int(choice) if choice else 0
selected_model_id = model_configs[idx]["id"]
except (ValueError, IndexError):
selected_model_id = model_configs[0]["id"]
print(f"✅ 主模型设定为: {selected_model_id}")
# 8. 写入配置
config_path = get_config_path()
print(f"\n💾 正在保存配置到 {config_path} ...")
config = load_config(config_path)
backup_config(config_path)
# 构造 Provider 块
provider_block = {
"baseUrl": base_url,
"apiKey": api_key,
"api": template["api_type"],
"models": model_configs
}
# 更新 providers
if "models" not in config:
config["models"] = {}
if "providers" not in config["models"]:
config["models"]["providers"] = {}
config["models"]["providers"][provider_id] = provider_block
# 更新 Agents 默认设置
if "agents" not in config:
config["agents"] = {}
if "defaults" not in config["agents"]:
config["agents"]["defaults"] = {}
defaults = config["agents"]["defaults"]
defaults["model"] = {"primary": f"{provider_id}/{selected_model_id}"}
# 清理并重建模型映射
if "models" not in defaults:
defaults["models"] = {}
prefix = f"{provider_id}/"
# 删除旧的同提供商映射
keys_to_del = [k for k in defaults["models"].keys() if k.startswith(prefix)]
for k in keys_to_del:
del defaults["models"][k]
# 添加新映射
for mc in model_configs:
key = f"{prefix}{mc['id']}"
# 如果是主模型，给个别名标记（可选逻辑，视 OpenClaw 具体需求）
defaults["models"][key] = {"alias": provider_id} if mc["id"] == selected_model_id else {}
if save_config(config, config_path):
print(f"\n✅ 配置成功！")
else:
print("\n❌ 保存失败。")
sys.exit(1)
# 9. 总结
print("\n" + "=" * 70)
print("📋 配置摘要")
print("=" * 70)
print(f"✓ 提供商 ID: {provider_id}")
print(f"✓ 端点: {base_url}")
print(f"✓ 已配置模型数: {len(model_configs)}")
print(f"✓ 主模型: {provider_id}/{selected_model_id}")
print(f"\n🚀 下一步操作:")
print(f" 1. 验证配置: openclaw gateway --check (如果支持)")
print(f" 2. 启动网关: openclaw gateway")
print(f" 3. 测试对话: openclaw agent --agent main -m \"Hello from cloud\" --new")
print("=" * 70)
if __name__ == "__main__":
try:
main()
except KeyboardInterrupt:
print("\n\n❌ 用户取消操作")
sys.exit(1)
except Exception as e:
print(f"\n❌ 发生未捕获错误: {e}")
import traceback
traceback.print_exc()
sys.exit(1)
运行方式
chmod +x setup_cloud_models.py
