openclaw-starter-guideOpenClaw 小白养成手册。从零开始搭建多 Agent AI 助手系统的完整指南，包含免费起步方案（Qwen 零成本）、进阶方案（MiniMax Coding Plan ¥49/月）、旗舰外援（SiliconFlow/NewCLI 按需调用）。涵盖模型策略、Fallback 链设计、额度管理、常见故障排查。适合首次部署 OpenClaw 或想优化模型成本的用户。当用户说"怎么开始"、"新手指南"、"省钱方案"、"模型怎么选"、"fallback 怎么配"时使用此 skill。
Install via ClawdBot CLI:
clawdbot install jooey/openclaw-starter-guide从零到多 Agent AI 系统的完整指南。
OpenClaw 的模型策略就像组建一支球队：
🆓 免费球员（Qwen） → 保底，确保系统永远在线
💰 包月主力（MiniMax） → 日常主力，性价比最高
🚀 外援专家（旗舰模型） → 关键时刻上场，按需付费目标：用最少的钱，让 AI 助手 24/7 在线、随时可用。
clawhub CLI）# 安装 ClawHub CLI（用于安装 skills）
npm i -g clawhub目标：¥0 成本让系统跑起来。
| 来源 | 模型 | 特点 | 安装 Skill |
|------|------|------|------------|
| Qwen Portal | qwen-portal/coder-model | OAuth 免费，128K context | 内置 |
| SiliconFlow | Qwen/Qwen3-8B | API Key 免费，无限调用 | clawhub install add-siliconflow-provider |
| SiliconFlow | DeepSeek-R1-0528-Qwen3-8B | 免费推理模型 | 同上 |
只需一个 provider 就能开始：
{
"models": {
"providers": {
"siliconflow": {
"baseUrl": "https://api.siliconflow.cn/v1",
"apiKey": "<YOUR_KEY>",
"api": "openai-completions",
"models": [
{
"id": "Qwen/Qwen3-8B",
"name": "Qwen3 8B (Free)",
"reasoning": false,
"input": ["text"],
"cost": {"input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0},
"contextWindow": 32768,
"maxTokens": 8192
}
]
}
}
},
"agents": {
"defaults": {
"model": {
"primary": "siliconflow/Qwen/Qwen3-8B"
}
}
}
}注册 SiliconFlow：https://cloud.siliconflow.cn/i/ihj5inat （新用户送 ¥14）
目标：¥49/月获得企业级体验。
当免费模型不够用时，上 MiniMax Coding Plan：
| 项目 | 值 |
|------|------|
| 价格 | ¥49/月 |
| 额度 | 1500 次/5小时窗口（约 7200 次/天） |
| 模型 | MiniMax M2.1（200K context） |
| 质量 | 对标 Claude Sonnet / GPT-4o |
openai-completions 协议即插即用clawhub install add-minimax-provider详细配置见：add-minimax-provider
{
"agents": {
"defaults": {
"model": {
"primary": "minimax/MiniMax-M2.1",
"fallbacks": [
"siliconflow/Qwen/Qwen3-8B"
]
}
}
}
}关键：免费模型放在 fallback 里！MiniMax 挂了或额度用完时，自动切到免费模型，系统不会停。
目标：关键任务用最强模型，按需付费。
| 场景 | 推荐模型 | 来源 | 价格 |
|------|----------|------|------|
| 复杂推理 | DeepSeek R1 | SiliconFlow | ¥4/¥16 per M |
| 代码重构 | Qwen3 Coder 480B | SiliconFlow | ¥8/¥16 per M |
| 顶级对话 | Kimi K2.5 | SiliconFlow | ¥4/¥21 per M |
| 长文档 | Claude Opus | NewCLI | 按额度 |
| 多模态 | Gemini 3 Pro | NewCLI | 按额度 |
clawhub install add-siliconflow-provider # DeepSeek/Qwen/Kimi
clawhub install add-newcli-provider # Claude/GPT/Gemini不需要改默认配置。需要时用 /model 命令临时切换：
/model sf-kimi # 切到 Kimi K2.5
/model sf-coder-480b # 切到 Qwen3 Coder 480B
/model claude-opus # 切到 Claude Opus
/model Minimax # 切回 MiniMax（默认）Fallback 链是 OpenClaw 的生命线——主模型挂了，自动尝试下一个。
第1优先：minimax/MiniMax-M2.1 (API Key 包月主力)
↓ 如果额度用完或 API 故障
第2优先：minimax-portal/MiniMax-M2.1 (OAuth 免费额度)
↓ 如果 OAuth 也不可用
第3优先：siliconflow/Qwen/Qwen3-8B (免费兜底)
↓ 如果 SiliconFlow 也挂了
第4优先：qwen-portal/coder-model (OAuth 免费)
↓ 如果都挂了
第5优先：deepseek/deepseek-chat (便宜)
↓ 最后的最后
第6优先：newcli/claude-haiku (贵但稳)"fallbacks": [
"minimax-portal/MiniMax-M2.1",
"siliconflow/Qwen/Qwen3-8B",
"qwen-portal/coder-model",
"deepseek/deepseek-chat",
"newcli/claude-haiku-4-5-20251001"
]1500 次/5小时滑动窗口
每次调用时倒算前 5 小时消耗
每天理论上限约 7200 次⚠️ 注意：MiniMax 额度查询 API (/coding_plan/remains) 的数据可能不准确（窗口切换后不及时刷新）。判断额度是否可用，最可靠的方法是发一个真实测试请求。
问题：白天集中使用，凌晨浪费额度。
解决：用 cron 安排夜间自动任务：
01:00 代码质量扫描
02:00 文档评审
03:00 TODO 整理
04:00 系统健康巡检
06:00 学习材料准备
07:00 早间简报🔴 重要：额度查询 API 数据不可信！
MiniMax 的额度查询 API 存在惰性更新问题——窗口切换后如果没有新调用，计数器不会刷新，返回的数字可能是上一窗口的残留数据。
推荐做法：不看数字，只看真实请求能不能通。
# 判断额度是否可用的唯一可靠方法
curl -s https://api.minimaxi.com/v1/chat/completions \
-H "Authorization: Bearer <API_KEY>" \
-H "Content-Type: application/json" \
-d '{"model":"MiniMax-M2.1","messages":[{"role":"user","content":"test"}],"max_tokens":3}'
# 返回 choices → 可用 | 返回 429 → 额度耗尽配合 OpenClaw cron 定期验证即可，无需频繁轮询 API 数字。
main (默认 Agent) → 处理所有事务main → 日常对话、任务协调
coder → 代码开发（用编码专用模型）
assistant → 日程、邮件、提醒
config-bot → 系统运维（用高级模型）| Agent 类型 | 推荐模型 | 理由 |
|-----------|----------|------|
| 日常对话 | MiniMax M2.1 | 包月，不心疼 |
| 代码开发 | MiniMax M2.1 / sf-coder-30b | 编码质量好 |
| 系统运维 | Claude Opus | 需要最高可靠性 |
| 轻量任务 | Qwen3-8B (免费) | 简单任务不浪费 |
# 1. 检查系统状态
openclaw doctor
# 2. 查看错误日志
tail -20 ~/.openclaw/logs/gateway.err.log
# 3. 测试模型可用性
curl -s '<BASE_URL>/chat/completions' \
-H 'Authorization: Bearer <KEY>' \
-H 'Content-Type: application/json' \
-d '{"model":"<MODEL>","messages":[{"role":"user","content":"test"}],"max_tokens":5}'sessions_list 看 totalTokens/contextTokens# 恢复备份
cp ~/.openclaw/openclaw.json.backup.<TIMESTAMP> ~/.openclaw/openclaw.json
openclaw gateway restart⚠️ 牢记：agents.defaults.models. 只允许 alias 字段！一个非法字段 = 全面崩溃。
/model sf-qwen3-8b（免费）详细故障排查见：openclaw-troubleshooting（开发中）
| 方案 | 月费 | 每日可用次数 | 适合 |
|------|------|------------|------|
| 纯免费 | ¥0 | 无限（质量有限） | 个人试玩 |
| MiniMax 包月 | ¥49 | ~7200 | 日常使用 |
| MiniMax + SF 余额 | ¥49 + ¥50 | 7200 + 按需旗舰 | 进阶用户 |
| 全家桶 | ¥49 + ¥50 + NewCLI | 全模型覆盖 | 重度用户 |
| 技能 | 安装命令 | 说明 |
|------|----------|------|
| SiliconFlow | clawhub install add-siliconflow-provider | 98+ 模型，含免费 |
| MiniMax | clawhub install add-minimax-provider | ¥49/月包月方案 |
| NewCLI | clawhub install add-newcli-provider | Claude/GPT/Gemini |
Generated Mar 1, 2026
A small e-commerce store uses the free Qwen3-8B model for basic customer inquiries and order tracking, with MiniMax M2.1 as the primary model for handling complex refund requests or product recommendations. This setup ensures 24/7 support at minimal cost, leveraging fallback chains to maintain service during high traffic.
A freelance developer deploys OpenClaw with MiniMax M2.1 for coding tasks and project management, using SiliconFlow's free models for simple reminders and file organization. The system automates code reviews and schedule planning, optimizing the ¥49/month plan for professional use without overspending.
An online tutoring service uses OpenClaw with Qwen3-8B for answering basic student questions and MiniMax M2.1 for detailed explanations in math or science. Fallback chains ensure reliability during peak study hours, and flagship models like DeepSeek R1 are reserved for advanced problem-solving sessions.
A tech startup implements a multi-agent system with a main agent for team coordination, a coder agent for development tasks using MiniMax M2.1, and a config-bot with Claude Opus for critical system updates. This architecture balances cost efficiency with high reliability for scaling operations.
A marketing agency uses OpenClaw with MiniMax M2.1 for generating blog drafts and social media content, while free models handle scheduling and basic edits. Flagship models like Kimi K2.5 are invoked via commands for high-stakes client campaigns, ensuring quality without constant premium costs.
Offer a free tier using SiliconFlow's Qwen3-8B for basic AI assistance, with paid subscriptions starting at ¥49/month for MiniMax M2.1 access. Upsell to premium plans that include flagship model credits, targeting small businesses and individual users seeking cost-effective scalability.
Provide customized OpenClaw deployments with multi-agent architectures, using MiniMax M2.1 as the core model and fallback chains for reliability. Charge based on agent count and usage, with add-ons for flagship model integration and dedicated support, catering to mid-sized companies.
Offer consulting to help clients set up OpenClaw with optimal model strategies, including fallback design and cron jobs for even usage. Revenue comes from one-time setup fees and ongoing maintenance contracts, focusing on industries like e-commerce and education.
💬 Integration Tip
Start with the free SiliconFlow provider and a single agent, then gradually add MiniMax and fallback chains as needs grow to avoid upfront costs.
Transform AI agents from task-followers into proactive partners that anticipate needs and continuously improve. Now with WAL Protocol, Working Buffer, Autonomous Crons, and battle-tested patterns. Part of the Hal Stack 🦞
Use the ClawdHub CLI to search, install, update, and publish agent skills from clawdhub.com. Use when you need to fetch new skills on the fly, sync installed skills to latest or a specific version, or publish new/updated skill folders with the npm-installed clawdhub CLI.
Clawdbot documentation expert with decision tree navigation, search scripts, doc fetching, version tracking, and config snippets for all Clawdbot features
Interact with Moltbook social network for AI agents. Post, reply, browse, and analyze engagement. Use when the user wants to engage with Moltbook, check their feed, reply to posts, or track their activity on the agent social network.
OpenClaw CLI wrapper — gateway, channels, models, agents, nodes, browser, memory, security, automation.
MoltGuard — runtime security plugin for OpenClaw agents by OpenGuardrails. Helps users install, register, activate, and check the status of MoltGuard. Use wh...