编辑OpenClaw配置文件

1152 字

如何使用OpenRouter,才能充分发挥它整合各大模型的汇合作用?也就是,在OPENCLAW里,如何配置OpenRouter,在不同应用场景使用它整合的不同大模型?

[2026/3/17 17:49] fangxia_bot: 我来为你详细解释如何在OpenClaw中配置和使用OpenRouter来充分发挥其整合各大模型的作用。

OpenRouter在OpenClaw中的配置与使用指南

  1. OpenRouter的核心优势

OpenRouter是一个模型聚合平台,让你通过单一API访问多个主流AI模型(GPT-4、Claude、Gemini、Llama等),具有以下优势:

• 统一API:一套接口调用所有模型
• 成本优化:自动选择性价比最高的模型
• 负载均衡:避免单一模型过载
• 功能互补:不同任务使用最适合的模型

  1. OpenClaw中配置OpenRouter

步骤1:获取OpenRouter API密钥

  1. 访问 OpenRouter.ai (https://openrouter.ai/)
  2. 注册账号并登录
  3. 在Dashboard中创建API密钥
  4. 查看可用模型列表和定价

步骤2:在OpenClaw中配置

编辑OpenClaw配置文件

nano ~/.openclaw/openclaw.json

添加OpenRouter配置:

{
"providers": {
"openrouter": {
"apiKey": "你的OpenRouter_API密钥",
"baseURL": "https://openrouter.ai/api/v1",
"defaultModel": "openrouter/auto", // 自动选择
"models": {
"fast": "openrouter/google/gemini-2.0-flash-exp:free",
"smart": "openrouter/anthropic/claude-3.5-sonnet",
"creative": "openrouter/openai/gpt-4o",
"coding": "openrouter/meta-llama/llama-3.3-70b-instruct"
}
}
}
}

步骤3:设置环境变量(可选)

在~/.bashrc或~/.zshrc中添加

export OPENROUTER_API_KEY="你的API密钥"
export OPENCLAW_DEFAULT_MODEL="openrouter/auto"

  1. 不同场景的模型选择策略

场景1:日常对话和快速响应

{
"model": "openrouter/google/gemini-2.0-flash-exp:free",
"reason": "快速、免费、适合简单问答"
}

场景2:复杂分析和深度思考

{
"model": "openrouter/anthropic/claude-3.5-sonnet",
"reason": "推理能力强,适合复杂问题分析"
}

场景3:创意写作和内容生成

{
"model": "openrouter/openai/gpt-4o",
"reason": "创意丰富,语言表达优秀"
}

场景4:代码编写和技术问题

{
"model": "openrouter/meta-llama/llama-3.3-70b-instruct",
"reason": "代码能力强,技术理解深入"
}

场景5:成本敏感型任务

{
"model": "openrouter/google/gemini-2.0-flash-thinking-exp:free",
"reason": "免费且性能不错"
}

  1. 在OpenClaw会话中动态切换模型

方法1:使用会话命令

查看当前模型

/model status

切换到指定模型

/model use openrouter/anthropic/claude-3.5-sonnet

使用别名切换

/model use smart

重置为默认

/model reset

方法2:在技能中指定模型

在技能配置文件中指定:

在SKILL.md的metadata部分

metadata:
openclaw:
preferred_model: "openrouter/anthropic/claude-3.5-sonnet"
model_rules:
- when: "task contains '分析' or '思考'"
use: "openrouter/anthropic/claude-3.5-sonnet"
- when: "task contains '代码' or '编程'"
use: "openrouter/meta-llama/llama-3.3-70b-instruct"
- when: "task contains '创意' or '写作'"
use: "openrouter/openai/gpt-4o"

方法3:使用智能路由技能

让我搜索并安装一个智能模型路由技能:
[2026/3/17 17:49] fangxia_bot: 让我安装一个智能模型路由技能:
[2026/3/17 17:50] fangxia_bot: 让我查看这个技能的说明:
[2026/3/17 17:51] fangxia_bot: 太好了!这个模型路由技能正是你需要的。现在让我继续为你解释如何结合OpenRouter使用。

  1. OpenRouter与模型路由技能的集成

配置OpenRouter作为主要提供商

运行模型路由设置向导

cd ~/.openclaw/workspace/skills/model-router
python3 scripts/setup-wizard.py

在向导中选择:

  1. 添加OpenRouter作为提供商
  2. 输入OpenRouter API密钥
  3. 配置OpenRouter支持的模型映射

创建OpenRouter专用配置

{
"providers": {
"openrouter": {
"configured": true,
"models": [
"openrouter/auto",
"openrouter/anthropic/claude-3.5-sonnet",
"openrouter/openai/gpt-4o",
"openrouter/google/gemini-2.0-flash-exp",
"openrouter/meta-llama/llama-3.3-70b-instruct",
"openrouter/mistralai/mistral-large-2"
]
}
},
"task_mappings": {
"simple": "openrouter/google/gemini-2.0-flash-exp:free",
"coding": "openrouter/meta-llama/llama-3.3-70b-instruct",
"analysis": "openrouter/anthropic/claude-3.5-sonnet",
"creative": "openrouter/openai/gpt-4o",
"reasoning": "openrouter/anthropic/claude-3-opus"
}
}

  1. 实际应用场景示例

场景A:自动化工作流

1. 简单查询使用免费模型

sessions_spawn --task "查询天气" --model openrouter/google/gemini-2.0-flash-exp:free

2. 代码问题使用Llama

sessions_spawn --task "修复Python内存泄漏" --model openrouter/meta-llama/llama-3.3-70b-instruct

3. 战略分析使用Claude

sessions_spawn --task "分析市场趋势并制定策略" --model openrouter/anthropic/claude-3.5-sonnet

4. 创意内容使用GPT-4

sessions_spawn --task "写一篇关于AI未来的文章" --model openrouter/openai/gpt-4o

场景B:成本优化策略

智能路由脚本示例

import subprocess
import json

def route_task(task_description, budget="low"):
"""根据预算智能路由任务"""

model_map = {
    "low": {
        "simple": "openrouter/google/gemini-2.0-flash-exp:free",
        "medium": "openrouter/meta-llama/llama-3.1-8b-instruct",
        "complex": "openrouter/anthropic/claude-3-haiku"
    },
    "medium": {
        "simple": "openrouter/anthropic/claude-3-haiku",
        "medium": "openrouter/anthropic/claude-3.5-sonnet",
        "complex": "openrouter/openai/gpt-4o-mini"
    },
    "high": {
        "simple": "openrouter/anthropic/claude-3.5-sonnet",
        "medium": "openrouter/openai/gpt-4o",
        "complex": "openrouter/anthropic/claude-3-opus"
    }
}

# 分析任务复杂度
complexity = analyze_complexity(task_description)

# 选择模型
model = model_map[budget][complexity]

# 执行任务
cmd = f'sessions_spawn --task "{task_description}" --model {model}'
subprocess.run(cmd, shell=True)
  1. 高级配置:动态模型选择

创建智能路由规则

~/.openclaw/rules/model_routing.yaml

rules:
- name: "代码相关任务"
conditions:
- task_contains: ["代码", "编程", "bug", "修复", "开发"]
action:
model: "openrouter/meta-llama/llama-3.3-70b-instruct"
timeout: 300

  • name: "数据分析任务"
    conditions:

    • task_contains: ["分析", "统计", "数据", "报表", "趋势"]
      action:
      model: "openrouter/anthropic/claude-3.5-sonnet"
      timeout: 600
  • name: "创意写作任务"
    conditions:

    • task_contains: ["写作", "创意", "文章", "故事", "营销"]
      action:
      model: "openrouter/openai/gpt-4o"
      timeout: 480
  • name: "快速查询任务"
    conditions:

    • task_length: "< 50"
    • not_contains: ["分析", "代码", "写作"]
      action:
      model: "openrouter/google/gemini-2.0-flash-exp:free"
      timeout: 60
  • 监控和优化

查看使用统计

查看各模型使用情况

openclaw status --providers

查看成本统计

openclaw billing --period monthly

查看性能指标

openclaw metrics --model-response-times

优化建议

  1. 定期审查:每月检查各模型的使用效果和成本
  2. A/B测试:对相同任务使用不同模型比较结果
  3. 成本警报:设置月度预算警报
  4. 性能监控:跟踪响应时间和准确率

  5. 故障排除

常见问题

  1. API密钥无效# 重新配置OpenRouter
    openclaw config set openrouter.apiKey "新密钥"
  2. 模型不可用# 查看可用模型
    curl -H "Authorization: Bearer $OPENROUTER_API_KEY" \
    https://openrouter
    [2026/3/17 17:49] fangxia_bot: 我来为你详细解释如何在OpenClaw中配置和使用OpenRouter来充分发挥其整合各大模型的作用。

OpenRouter在OpenClaw中的配置与使用指南

  1. OpenRouter的核心优势

OpenRouter是一个模型聚合平台,让你通过单一API访问多个主流AI模型(GPT-4、Claude、Gemini、Llama等),具有以下优势:

• 统一API:一套接口调用所有模型
• 成本优化:自动选择性价比最高的模型
• 负载均衡:避免单一模型过载
• 功能互补:不同任务使用最适合的模型

  1. OpenClaw中配置OpenRouter

步骤1:获取OpenRouter API密钥

  1. 访问 OpenRouter.ai (https://openrouter.ai/)
  2. 注册账号并登录
  3. 在Dashboard中创建API密钥
  4. 查看可用模型列表和定价

步骤2:在OpenClaw中配置

编辑OpenClaw配置文件

nano ~/.openclaw/openclaw.json

添加OpenRouter配置:

{
"providers": {
"openrouter": {
"apiKey": "你的OpenRouter_API密钥",
"baseURL": "https://openrouter.ai/api/v1",
"defaultModel": "openrouter/auto", // 自动选择
"models": {
"fast": "openrouter/google/gemini-2.0-flash-exp:free",
"smart": "openrouter/anthropic/claude-3.5-sonnet",
"creative": "openrouter/openai/gpt-4o",
"coding": "openrouter/meta-llama/llama-3.3-70b-instruct"
}
}
}
}

步骤3:设置环境变量(可选)

在~/.bashrc或~/.zshrc中添加

export OPENROUTER_API_KEY="你的API密钥"
export OPENCLAW_DEFAULT_MODEL="openrouter/auto"

  1. 不同场景的模型选择策略

场景1:日常对话和快速响应

{
"model": "openrouter/google/gemini-2.0-flash-exp:free",
"reason": "快速、免费、适合简单问答"
}

场景2:复杂分析和深度思考

{
"model": "openrouter/anthropic/claude-3.5-sonnet",
"reason": "推理能力强,适合复杂问题分析"
}

场景3:创意写作和内容生成

{
"model": "openrouter/openai/gpt-4o",
"reason": "创意丰富,语言表达优秀"
}

场景4:代码编写和技术问题

{
"model": "openrouter/meta-llama/llama-3.3-70b-instruct",
"reason": "代码能力强,技术理解深入"
}

场景5:成本敏感型任务

{
"model": "openrouter/google/gemini-2.0-flash-thinking-exp:free",
"reason": "免费且性能不错"
}

  1. 在OpenClaw会话中动态切换模型

方法1:使用会话命令

查看当前模型

/model status

切换到指定模型

/model use openrouter/anthropic/claude-3.5-sonnet

使用别名切换

/model use smart

重置为默认

/model reset

方法2:在技能中指定模型

在技能配置文件中指定:

在SKILL.md的metadata部分

metadata:
openclaw:
preferred_model: "openrouter/anthropic/claude-3.5-sonnet"
model_rules:
- when: "task contains '分析' or '思考'"
use: "openrouter/anthropic/claude-3.5-sonnet"
- when: "task contains '代码' or '编程'"
use: "openrouter/meta-llama/llama-3.3-70b-instruct"
- when: "task contains '创意' or '写作'"
use: "openrouter/openai/gpt-4o"

方法3:使用智能路由技能

让我搜索并安装一个智能模型路由技能:
[2026/3/17 17:49] fangxia_bot: 让我安装一个智能模型路由技能:
[2026/3/17 17:50] fangxia_bot: 让我查看这个技能的说明:
[2026/3/17 17:51] fangxia_bot: 太好了!这个模型路由技能正是你需要的。现在让我继续为你解释如何结合OpenRouter使用。

  1. OpenRouter与模型路由技能的集成

配置OpenRouter作为主要提供商

运行模型路由设置向导

cd ~/.openclaw/workspace/skills/model-router
python3 scripts/setup-wizard.py

在向导中选择:

  1. 添加OpenRouter作为提供商
  2. 输入OpenRouter API密钥
  3. 配置OpenRouter支持的模型映射

创建OpenRouter专用配置

{
"providers": {
"openrouter": {
"configured": true,
"models": [
"openrouter/auto",
"openrouter/anthropic/claude-3.5-sonnet",
"openrouter/openai/gpt-4o",
"openrouter/google/gemini-2.0-flash-exp",
"openrouter/meta-llama/llama-3.3-70b-instruct",
"openrouter/mistralai/mistral-large-2"
]
}
},
"task_mappings": {
"simple": "openrouter/google/gemini-2.0-flash-exp:free",
"coding": "openrouter/meta-llama/llama-3.3-70b-instruct",
"analysis": "openrouter/anthropic/claude-3.5-sonnet",
"creative": "openrouter/openai/gpt-4o",
"reasoning": "openrouter/anthropic/claude-3-opus"
}
}

  1. 实际应用场景示例

场景A:自动化工作流

1. 简单查询使用免费模型

sessions_spawn --task "查询天气" --model openrouter/google/gemini-2.0-flash-exp:free

2. 代码问题使用Llama

sessions_spawn --task "修复Python内存泄漏" --model openrouter/meta-llama/llama-3.3-70b-instruct

3. 战略分析使用Claude

sessions_spawn --task "分析市场趋势并制定策略" --model openrouter/anthropic/claude-3.5-sonnet

4. 创意内容使用GPT-4

sessions_spawn --task "写一篇关于AI未来的文章" --model openrouter/openai/gpt-4o

场景B:成本优化策略

智能路由脚本示例

import subprocess
import json

def route_task(task_description, budget="low"):
"""根据预算智能路由任务"""

model_map = {
    "low": {
        "simple": "openrouter/google/gemini-2.0-flash-exp:free",
        "medium": "openrouter/meta-llama/llama-3.1-8b-instruct",
        "complex": "openrouter/anthropic/claude-3-haiku"
    },
    "medium": {
        "simple": "openrouter/anthropic/claude-3-haiku",
        "medium": "openrouter/anthropic/claude-3.5-sonnet",
        "complex": "openrouter/openai/gpt-4o-mini"
    },
    "high": {
        "simple": "openrouter/anthropic/claude-3.5-sonnet",
        "medium": "openrouter/openai/gpt-4o",
        "complex": "openrouter/anthropic/claude-3-opus"
    }
}

# 分析任务复杂度
complexity = analyze_complexity(task_description)

# 选择模型
model = model_map[budget][complexity]

# 执行任务
cmd = f'sessions_spawn --task "{task_description}" --model {model}'
subprocess.run(cmd, shell=True)
  1. 高级配置:动态模型选择

创建智能路由规则

~/.openclaw/rules/model_routing.yaml

rules:
- name: "代码相关任务"
conditions:
- task_contains: ["代码", "编程", "bug", "修复", "开发"]
action:
model: "openrouter/meta-llama/llama-3.3-70b-instruct"
timeout: 300

  • name: "数据分析任务"
    conditions:

    • task_contains: ["分析", "统计", "数据", "报表", "趋势"]
      action:
      model: "openrouter/anthropic/claude-3.5-sonnet"
      timeout: 600
  • name: "创意写作任务"
    conditions:

    • task_contains: ["写作", "创意", "文章", "故事", "营销"]
      action:
      model: "openrouter/openai/gpt-4o"
      timeout: 480
  • name: "快速查询任务"
    conditions:

    • task_length: "< 50"
    • not_contains: ["分析", "代码", "写作"]
      action:
      model: "openrouter/google/gemini-2.0-flash-exp:free"
      timeout: 60
  • 监控和优化

查看使用统计

查看各模型使用情况

openclaw status --providers

查看成本统计

openclaw billing --period monthly

查看性能指标

openclaw metrics --model-response-times

优化建议

  1. 定期审查:每月检查各模型的使用效果和成本
  2. A/B测试:对相同任务使用不同模型比较结果
  3. 成本警报:设置月度预算警报
  4. 性能监控:跟踪响应时间和准确率

  5. 故障排除

常见问题

  1. API密钥无效# 重新配置OpenRouter
    openclaw config set openrouter.apiKey "新密钥"
  2. 模型不可用# 查看可用模型
    curl -H "Authorization: Bearer $OPENROUTER_API_KEY" \
    https://openrouter
    [2026/3/17 17:51] fangxia_bot: .ai/api/v1/models
  3. 响应超时# 增加超时时间
    model_config:
    timeout: 120 # 秒
    retry_count: 3

  4. 最佳实践总结

  5. 分层使用:
    • 简单任务 → 免费/低成本模型
    • 中等任务 → 性价比模型
    • 复杂任务 → 高性能模型

  6. 智能路由:
    • 使用模型路由技能自动选择
    • 根据任务类型动态切换
    • 考虑成本和性能平衡
  7. 监控优化:
    • 定期分析使用数据
    • 调整模型映射规则
    • 优化成本效益比
  8. 备份策略:
    • 配置多个备用模型
    • 设置故障转移机制
    • 保持API密钥更新

下一步行动建议:

  1. 立即配置:# 1. 获取OpenRouter API密钥
    [2026/3/17 17:51] fangxia_bot: # 2. 安装模型路由技能
    skillhub install model-router

3. 运行设置向导

cd ~/.openclaw/workspace/skills/model-router
python3 scripts/setup-wizard.py

4. 测试路由功能

python3 scripts/classify_task.py "帮我写一段Python代码"
2. 渐进实施:
• 先从简单任务开始测试
• 逐步增加复杂任务
• 根据效果调整配置
3. 长期优化:
• 每月审查使用报告
• 根据需求调整模型策略
• 探索新模型和功能

这样配置后,OpenClaw就能充分发挥OpenRouter整合各大模型的优势,在不同应用场景自动选择最合适的模型,实现成本、性能和功能的完美平衡。

你需要我帮你执行具体的配置步骤吗?

我已经为你整理了OpenRouter支持的主要大模型全名列表。这个列表包含了19个类别,超过100个模型,涵盖了所有主流AI提供商。

📋 OpenRouter主要大模型全名总结

核心要点:

  1. 主要提供商系列:

• OpenAI系列 (16个模型):GPT-5.1到GPT-5.4的各种版本
• Anthropic系列 (3个模型):Claude Opus 4.5/4.6, Claude Sonnet 4.6
• Google系列 (7个模型):Gemini 3.x 各种预览版
• Meta系列 (2个模型):Llama 3.3 70B
• Mistral AI系列 (7个模型):Mistral Large, Ministral, Devstral等
• DeepSeek系列 (2个模型):DeepSeek V3.2
• Qwen系列 (9个模型):Qwen 3.5 各种规模
• xAI系列 (3个模型):Grok 4.x 各种版本

  1. 特色模型:

• openrouter/auto - 智能自动路由(强烈推荐)
• openrouter/free - 免费模型
• 各系列的免费版本(带 :free 后缀)

  1. 按任务类型推荐:

• 通用对话:openrouter/auto, openai/gpt-5.2-chat
• 代码编程:openai/gpt-5.1-codex, deepseek/deepseek-v3.2
• 复杂推理:anthropic/claude-opus-4.6, openai/gpt-5.4-pro
• 中文任务:qwen/qwen3.5-27b, z-ai/glm-4.7
• 快速响应:openrouter/free, google/gemini-3.1-flash-lite-preview

在OpenClaw中的使用方式:

智能路由(推荐)

sessions_spawn --task "你的任务" --model openrouter/auto

指定模型

sessions_spawn --task "写代码" --model openai/gpt-5.1-codex
sessions_spawn --task "深度分析" --model anthropic/claude-opus-4.6
sessions_spawn --task "中文写作" --model qwen/qwen3.5-27b

免费使用

sessions_spawn --task "简单查询" --model openrouter/free

最佳实践建议:

  1. 日常使用:优先使用 openrouter/auto,让系统智能选择
  2. 专业任务:根据任务类型选择专门优化的模型
  3. 成本控制:简单任务使用免费或低成本模型
  4. 中文优化:中文任务优先选择Qwen、GLM等中文优化模型
  5. 定期更新:OpenRouter会不断添加新模型,建议定期查看

这个列表已经保存到 /tmp/openrouter_models.md,你可以随时参考。需要我帮你配置特定的模型路由规则吗?